* Makefile.am: Make 'make tarball-and-all' run the
'check-self-compare' target as part of the release process.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
It's useful to be able to force the build system into avoiding the use
of rpm 4.15 version or higher. That version of RPM is the one that
supports RPMs from Fedora 31 or higher. Those RPMs use the zstd
compression scheme. Prior to Fedora 31, RPM were not using the zstd
compression scheme. So, systems with rpm version lower than 4.15
cannot deal with RPMs coming from Fedora 31 or higher. So on those
systems, some regression tests of libabigail who use RPMs from Fedora
33 will fail.
With this patch, one can use the --disable-rpm415 option of the
configure script to prevent those tests from running on those pre 4.15
rpm systems.
* configure: Introduce the --{en, dis}able-rpm415 option.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
Sometimes a DW_TAG_partial_unit (imported via a DW_TAG_imported_unit)
has no children node. That means the the DW_TAG_partial_unit has no
sub-tree.
In those cases, the dwarf-reader crashes when
ctxt.build_die_parent_relations_under tries to record the point at
which the sub-tree (which is non-existent here) of the partial unit is
imported by the DW_TAG_imported_unit DIE.
This patch avoids crashing in those cases. As this problem is
reported on libclang-cpp.so (on Fedora 33) which takes "abidw --abidiff"
five hours and ~ 20GB of ram to complete at this point (on a power7 box)
this patch doesn't have a regression test attached.
* src/abg-dwarf-reader.cc (die_has_children): Define new static
function.
(read_context::build_die_parent_relations_under): Do not try to
instantiate an imported_unit_point type for an imported unit with
no children node.
(imported_unit_point::imported_unit_point): Assert that the
imported die has a sub-tree.
(imported_unit_point::imported_unit_point): Remove useless spaces.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
* src/abg-writer.cc (write_array_subrange_type): Fix off-by-one
error in assertion.
* src/abg-dwarf-reader.cc (build_subrange_type): Assert the length
of the array complies with its bounds.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
Now that several subtle causes of spurious ABI change report when
comparing a binary against its own ABIXML have been addressed, this
patch makes 'abipkgdiff --self-check' to fail on any ABI change
reported. That is, harmless changes are not ignored anymore.
* tools/abipkgdiff.cc (compare_to_self): Report *any* ABI change.
Not just the "net" changes.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
Sometimes, "abidw --abidiff <binary>" would pass while
"abidw <binary> > abi; abidiff <binary> abi" would fail.
This is because "abidw --abidiff" emits an error only when the
comparison between the binary and its ABIXML representation yields and
incompatible change.
Now that many subtle causes of spurious ABI change report emitted when
comparing a binary against its own ABIXML have been fixed, this patch
makes it so that *any* change would make abidw --abidiff to emit an
error.
* tools/abidw.cc (load_corpus_and_write_abixml): Emit an error
when comparing the binary to its ABIXML representation yields any
change.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
In the DWARF debug info, a C++ class can be represented by pieces
throughout a given binary.
In this picture, a given virtual member function can be represented
several times; each time in one different piece of the C++ class. In
a given piece of the class, a virtual member function can be
represented with its ELF symbol set. In another one, the same virtual
member function can be represented without that ELF symbol set. And
there can also be pieces of the class that don't have a given virtual
function.
To handle this, the DWARF reader constructs one class from all its
pieces scattered around. It adds each virtual member function to the
class as it comes across them while scanning the DWARF.
Then there is a pass at the end of the process that sets ELF symbols
to the (virtual) member functions that need it.
The problem with that pass is that it sometimes sets the same ELF
symbol to more than one virtual member function. Those virtual member
functions all have the same mangled name that correspond to the ELF
symbol; but only one of them is meant to be associated with the ELF
symbol. In essence, that one is the one that is exported by the ELF
binary.
This patch teaches the pass that sets the ELF symbols of function to
avoid setting the same ELF symbol to more than one function.
The patch also avoids build_function_decl to set symbol to a function
if that symbol was already set to an existing function.
This patch thus fixes a class of issues what arise when comparing a
binary against its own ABIXML representation. Those several functions
having the same ELF symbol would cause spurious changes in that
context.
* src/abg-dwarf-reader.cc
(read_context::symbol_already_belongs_to_a_function): Define new
member function.
(read_context::fixup_functions_with_no_symbols): Use the new
symbol_already_belongs_to_a_function function to avoid setting a
symbol that already belongs to a function.
* tests/data/test-read-dwarf/PR22122-libftdc.so.abi: Adjust.
* tests/data/test-read-dwarf/test10-pr18818-gcc.so.abi: Likewise.
* tests/data/test-read-dwarf/test16-pr18904.so.abi: Likewise.
* tests/data/test-read-dwarf/test22-pr19097-libstdc++.so.6.0.17.so.abi:
Likewise.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
When reading an anonymous declaration-only class from ABIXML
libabigail forgets to set the is-anonymous class. This leads to
spurious change reports when comparing a binary against its ABIXML
representation. Fixed thus.
Note that this doesn't yet impact any regression test but is useful
for a coming patch that will make abidw --abidiff to emit an error for
all ABI changes, not just the hard incompatible ones. Without this
change, that coming patch will make runtestreaddwarf to fail.
* src/abg-reader.cc (build_class_decl): Set the is-anonymous flag
when reading a decl-only class.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
There are two views for internal pretty representation of anonymous
classes.
1/ When we look at the anonymous class itself, we use its 'flat
representation' i.e:
'class {int blah; char bleh;}'
2/ When we look at a pointer or a reference to the anonymous class we
use its generic anonymous internal name, i.e:
'__anonymous_struct__*'
As a general rule, libabigail always use the keyword 'class' to prefix
the name of classes for internal purposes, independent from the fact
that the type is a struct or a class. That is a pre-requisite to be
able to canonicalize classes and structs together. In other words, if
a class and a struct are structurally equal, they are going to be
considered equivalent by the canonicalization process.
Currently however, in the view 1/ of the pretty representation of
anonymous classes, a struct and a class will have different
representations. For instance, and empty anonymous struct would be
represented as 'struct {}', whereas an empty anonymous class would be
represented as 'class {}'. This prevents these two be considered
equivalent by the canonicalization process. This leads to spurious
change reports later down the road.
In the view 2/ we have a similar but different problem: the qualified
names of the anonymous classes are not taken into account when
representing pointer or references to said anonymous classes. Only
their unqualified generic anonymous internal names are taken into
account in the representation. This leads to pointers/references to
anonymous classes being wrongly considered equivalent even when they
belong to different namespaces.
This patch corrects the issues related to both views 1/ and 2/. It
should make libabigail correctly consider some anonymous classes as
equivalent (view 1) and correctly consider pointers/references to
anonymous classes as different when they belong to different
namespaces (view 2).
A number of reference tests are adjusted accordingly.
* include/abg-fwd.h (get_class_or_union_flat_representation):
Introduce an "internal" parameter.
* src/abg-ir.cc (get_class_or_union_flat_representation):
Introduce an "internal" parameter. In the flat representation of
a class for internal purposes, always use the prefix "class" even
if this is a struct.
(get_type_name): To build an internal name for a
reference or pointer to an anonymous type, consider the namespace
name of said type.
(equals): In the overload for decl_base, take the namespace name
of anonymous decls into account when comparing them.
({var_decl, union_decl}::get_pretty_representation): Adjust calls
to get_class_or_union_flat_representation to pass a proper
"internal" argument.
* src/abg-default-reporter.cc (default_reporter::report): Adjust
the call to get_class_or_union_flat_representation to pass an
"internal" argument set to 'false'.
* tests/data/test-annotate/libtest23.so.abi: Adjust.
* tests/data/test-read-dwarf/PR22122-libftdc.so.abi: Likewise.
* tests/data/test-read-dwarf/libtest23.so.abi: Likewise.
* tests/data/test-read-dwarf/test10-pr18818-gcc.so.abi: Likewise.
* tests/data/test-read-dwarf/test11-pr18828.so.abi: Likewise.
* tests/data/test-read-dwarf/test12-pr18844.so.abi: Likewise.
* tests/data/test-read-dwarf/test9-pr18818-clang.so.abi: Likewise.
* tests/data/test-read-dwarf/test-libandroid.so.abi: Likewise.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
Libabigail goes a long way to resolve declaration-only classes
to their definitions when it's possible.
The ABIXML writer however sometimes forgets to emit the definition of
such declarations that have been "resolved".
Later, when the binary is compared to its own ABIXML representation,
the reporting engine thus reports that the definition is lost.
This patch fixes that.
* src/abg-writer.cc (write_class_decl, write_union_decl): Get the
definition of the declaration if it exists and emit that.
* tests/data/test-read-dwarf/test13-pr18894.so.abi: Adjust.
* tests/data/test-read-dwarf/test15-pr18892.so.abi: Likewise.
* tests/data/test-read-dwarf/test21-pr19092.so.abi: Likewise.
* tests/data/test-annotate/test13-pr18894.so.abi: Likewise.
* tests/data/test-annotate/test15-pr18892.so.abi: Likewise.
* tests/data/test-annotate/test21-pr19092.so.abi: Likewise.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
When serializing subrange bounds to ABIXML today the writer omits the
lower and upper bounds. Only the length is emitted. The reader thus
assumes that the lower bound of an array subrange is always 0. In
Fortran however, that is not true, as the lower bound is 1.
This patch instructs the writer to emits the lower bound whenever it's
different from zero. It also emits the upper bound in that case. The
reader is updated accordingly to take the lower and upper bounds into
account whenever they are present. It they are not, then the
lower bound is assumed to be zero and the upper bound is deduced from
the length, as was already the case until now.
* src/abg-reader.cc (build_subrange_type): Read lower-bound
attribute if present. Then try to read upper-bound attribute as
well. If this is not an infinite subrange assert that the length
must be equal to the difference between the bounds.
* src/abg-writer.cc (write_array_subrange_type): Write the
lower-bound if it's present and not zero. In that case, write the
upper-bound as well.
* tests/data/test-diff-pkg/hdf5-1.10.6-2.fc33.x86_64.rpm: Add new
binary test input.
* tests/data/test-diff-pkg/hdf5-debuginfo-1.10.6-2.fc33.x86_64.rpm:
Likewise.
* tests/data/test-diff-pkg/hdf5-1.10.6-2.fc33.x86_64.self-check-report-0.txt:
Add new reference output.
* tests/data/Makefile.am: Add the new test material above to
source distribution.
* tests/test-diff-pkg.cc (in_out_specs): Add the binary test input
to the set of --self-check tests.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
We read subrange length like an integer of type 'int" today. We
should read it as an uint64_t. Fixed thus.
* src/abg-reader.cc (build_subrange_type): Change the type of
length to uint64_t and read it using strtoull.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
Worker tasks that have compared the binaries of two packages are
sorted according to the sizes of the binaries they compared. The
tasks that compared bigger binaries come first. So the output of
these tasks is always sorted the same.
When two tasks have sorted binaries of the same size, let's sort them
by taking into account the lexicographic order of the binaries names.
* elf_size_is_greater: Take the name of the binaries into account
when their size is equal. Also, assert that all comparison tasks
have compared binaries.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
Instances of self_compare_task don't wrongly shadow task comparison
arguments anymore. Instead, they properly use the task comparison
arguments of the parent class compare_task.
So in the sorting function elf_size_is_greater, we now always have
proper elf binaries size. It follows that the sorting can now yield a
consistent result, always.
It thus appears that the reference output recorded for the
"self-comparison" test of runtestfedabipkgdiffpy3.sh was wrong. This
patch updates it.
* tests/data/test-fedabipkgdiff/test7-self-compare-from-fc23-dbus-glib-report-0.txt:
Adjust.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
The recent changes to abipkgdiff triggered some minor clang-tidy
diagnostics that I address in this patch.
- removed unused using statements
- cleaned up some mismatches between commented and actual parameter names
- removed shadowing data members from self_compare_task and construct
with base class constructor instead.
* tools/abipkgdiff.cc: remove unused using statements
(self_compare_task): remove shadowing members and delegate
construction to base class.
Signed-off-by: Matthias Maennich <maennich@google.com>
When GCC 10 artificially generates a translation unit, the path/name
of the translation unit as given by the DW_AT_name attribute of the
translation unit DIE is the string "<artificial>".
Libabigail expects that translation units have unique file paths so
having several artificially generated translation units like this one
with the same name makes hell break loose down the road.
This patch suffixes the name of artificial DIE with their die offset
number to have a unique path name for artificially generated
translation units.
* configure.ac: Detect if we are running on RPM >= 4.15. If yes,
then define the preprocessor macro RPM_4_15. If that macro is
defined then test-diff-pkg.cc can support RPMs from Fedora >= 31
as those are compressed with zstd. Earlier RPM versions don't
support that compression scheme.
* src/abg-dwarf-reader.cc (build_translation_unit_and_add_to_ir):
Suffix the offset of the translation unit to its name when that
name is "<artificial>".
* tests/data/test-diff-pkg/mesa-libGLU-9.0.1-3.fc33.x86_64.rpm:
New binary test input.
* tests/data/test-diff-pkg/mesa-libGLU-debuginfo-9.0.1-3.fc33.x86_64.rpm: Likewise.
* tests/data/test-diff-pkg/mesa-libGLU-9.0.1-3.fc33.x86_64.self-check-report-0.txt:
New reference output for the binary test input above.
* tests/data/Makefile.am: Add the new test inputs above to source
distribution.
* tests/test-diff-pkg.cc (in_out_specs): Add the binary test
inputs above to source distribution if we are running on an RPM
version >= 4.15.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
Now that the abipkgdiff program supports the --self-check option to
make it compare the binaries in an RPM against their own ABIXML
representation, make the --self-compare option of fedabipkgdiff use
the abipkgdiff --self-check.
* tools/fedabipkgdiff (abipkgdiff): If the user provides the
--self-compare options, generate the abipkgdiff command by using
the --self-check option.
(run_abipkgdiff): Each return value of the abipkgidiff runs can be
negative because they are unsigned values in essence, but as
python doesn't seem to have a unsigned integer type. So we need
to consider the max of the absolute value of the return codes
here.
* tests/data/test-fedabipkgdiff/test7-self-compare-from-fc23-dbus-glib-report-0.txt:
Adjust.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
After the report of
https://sourceware.org/bugzilla/show_bug.cgi?id=26769 it became
apparent that we want to be able compare a binary against its ABIXML
representation, including for cases where the binaries are embedded in
RPM packages.
This patches thus introduces the --self-check option to abipkgdiff so
that it can be invoked like this:
$ abipkgdiff --self-check --d1 libstdc++-debuginfo-10.2.1-8.fc34.x86_64.rpm --d1 gcc-debuginfo-10.2.1-8.fc34.x86_64.rpm libstdc++-10.2.1-8.fc34.x86_64.rpm
==== SELF CHECK SUCCEEDED for 'libstdc++.so.6.0.28'====
$
With this option, libabigail compares each binary in the RPM against
its own ABIXML representation.
This should hopefully help to write regression tests which have as
sole inputs the links to download the RPMs. It's also useful to ease
the process of reproducing the issue raised.
This option can now be used, for instance, by the libabigail-selfcheck
program over at https://pagure.io/libabigail-selfcheck.
* tools/abipkgdiff.cc (options::self_check): Define new data
member.
(options::options): Initialize it.
(display_usage): Add help string for the --self-check option.
(parse_command): Parse the new --self-check option.
(extract_deb): Add missing newline.
(compare): Remove useless white space.
(compare_to_self, self_compare_prepared_userspace_package)
(self_compare_prepared_package, compare_to_self): Add new static
functions.
(class self_compare_task): Add new class.
(prepare_package): Add a new overload that takes just one
parameter.
(elf_size_is_greater): Don't crash if the args are empty.
(main): If the --self-check option is given, make sure we have
just one package in argument. Use the new compare_to_self
function to handle the --self-check option.
* doc/manuals/abipkgdiff.rst: Add documentation for the new
--self-check option.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
The symptom of the issue at hand is that sometimes there can be types
missing from the abixml output. This happens when analysing some C++
code bases.
The core of the issue is the following. Support we have a type
"struct S" defined somewhere as:
struct S // #0
{
int dm1;
char dm2;
};
S s;
Suppose that in another translation unit, we have the class 'S' being
extended to add a member type to it:
struct S // #1
{
typedef int dm1_type;
};
typedef S::dm1_type Integer;
Integer something;
When emitting the abixml for the codebase, the definition of the
typedef S::dm1_type can be missing.
Note that in location #1, struct S is considered declaration-only.
It's definition is in another translation unit, in location #0.
So the abixml writer emits the 'struct S' defined in location #0, but
forgets to emit the 'struct S' in #1, which is indirectly used for the
sole purpose of using its member type S::dm1_type.
This patch emits the S::dm1_type type that is mistakenly forgotten
today.
Now that the "struct S" of #1 is also emitted, a tangent problem is
uncovered: S in #0 can be wrongly thought to be equivalent to S in #1,
for ABI purposes
This is because of an ODR-based optimization that is used for C++.
That is, the two struct S can be wrongly considered equivalent just
because they have the same name. Note that ODR means "One Definition Rule[1]"
This patch removes the ODR-based optimization and thus fixes many of
the issues uncovered by the previous changes.
The patch also uncovered that some non-static variables were sometimes wrongly
being added to the set of exported variables, while libabigail reads
corpora from abixml. The patch fixes this as well.
[1]: One Definition Rule: https://en.wikipedia.org/wiki/One_Definition_Rule
* include/abg-corpus.h (corpus::{record_canonical_type,
lookup_canonical_type}): Remove function declarations.
* src/abg-corpus-priv.h (corpus::priv::canonical_types_): Remove
data member.
* src/abg-corpus.cc (corpus::{record_canonical_type,
lookup_canonical_type}): Remove functions.
* src/abg-ir.cc (type_eligible_for_odr_based_comparison): Remove
static function.
(type_base::get_canonical_type_for): Don't perform the ODR-based
optimization for C++ anymore.
* src/abg-reader.cc
(read_context&::maybe_add_var_to_exported_decls): Don't add a
variable that hasn't been added to its scope. Otherwise, it means
we added a variable that wasn't yet properly constructed. Also
add a new overload for var_decl_sptr&.
(build_var_decl): Do not add the var to its the set of exported
declaration before we are sure it has been fully constructed and
added to the scope it belongs.
(build_class_decl): Only add *static* data members to the list of
exported declarations.
(handle_var_decl): A var decl seen here is a global variable
declaration. Add it to the list of exported declarations.
* src/abg-writer.cc (write_context::decl_only_type_is_emitted):
Constify parameter.
(write_translation_unit): Do not forget to emit referenced types
that were maybe not canonicalized. Also, avoid using noop_deleter
when it's not necessary.
(write_namespace_decl): Do not forget to emit canonicalized types
that are present in namespaces other than the global namespace.
* tests/runtestslowselfcompare.sh.in: New test that compares
libabigail.so against its own ABIXML representation.
* tests/Makefile.am: Add the new test runtestslowselfcompare.sh to
source distribution. This test is too slow to be run during the
course of 'make check'. It takes more than 5 minutes on my slow
box here. Rather, it can be run using 'make check-self-compare'.
I plan to run this before releases now.
* tests/data/test-annotate/libtest24-drop-fns-2.so.abi: Adjust.
* tests/data/test-annotate/libtest24-drop-fns.so.abi: Likewise.
* tests/data/test-annotate/test0.abi: Likewise.
* tests/data/test-annotate/test13-pr18894.so.abi: Likewise.
* tests/data/test-annotate/test14-pr18893.so.abi: Likewise.
* tests/data/test-annotate/test15-pr18892.so.abi: Likewise.
* tests/data/test-annotate/test17-pr19027.so.abi: Likewise.
* tests/data/test-annotate/test18-pr19037-libvtkRenderingLIC-6.1.so.abi:
Likewise.
* tests/data/test-annotate/test19-pr19023-libtcmalloc_and_profiler.so.abi:
Likewise.
* tests/data/test-annotate/test20-pr19025-libvtkParallelCore-6.1.so.abi:
Likewise.
* tests/data/test-annotate/test21-pr19092.so.abi: Likewise.
* tests/data/test-read-dwarf/PR22015-libboost_iostreams.so.abi:
Likewise.
* tests/data/test-read-dwarf/PR22122-libftdc.so.abi: Likewise.
* tests/data/test-read-dwarf/PR25042-libgdbm-clang-dwarf5.so.6.0.0.abi:
Likewise.
* tests/data/test-read-dwarf/PR26261/PR26261-exe.abi: Likewise.
* tests/data/test-read-dwarf/libtest24-drop-fns-2.so.abi:
Likewise.
* tests/data/test-read-dwarf/libtest24-drop-fns.so.abi: Likewise.
* tests/data/test-read-dwarf/test-libandroid.so.abi: Likewise.
* tests/data/test-read-dwarf/test0.abi: Likewise.
* tests/data/test-read-dwarf/test0.hash.abi: Likewise.
* tests/data/test-read-dwarf/test10-pr18818-gcc.so.abi: Likewise.
* tests/data/test-read-dwarf/test11-pr18828.so.abi: Likewise.
* tests/data/test-read-dwarf/test12-pr18844.so.abi: Likewise.
* tests/data/test-read-dwarf/test14-pr18893.so.abi: Likewise.
* tests/data/test-read-dwarf/test15-pr18892.so.abi: Likewise.
* tests/data/test-read-dwarf/test16-pr18904.so.abi: Likewise.
* tests/data/test-read-dwarf/test17-pr19027.so.abi: Likewise.
* tests/data/test-read-dwarf/test18-pr19037-libvtkRenderingLIC-6.1.so.abi:
Likewise.
* tests/data/test-read-dwarf/test19-pr19023-libtcmalloc_and_profiler.so.abi:
Likewise.
* tests/data/test-read-dwarf/test20-pr19025-libvtkParallelCore-6.1.so.abi:
Likewise.
* tests/data/test-read-dwarf/test21-pr19092.so.abi: Likewise.
* tests/data/test-read-dwarf/test22-pr19097-libstdc++.so.6.0.17.so.abi:
Likewise.
* tests/data/test-read-dwarf/test9-pr18818-clang.so.abi: Likewise.
* tests/data/test-read-write/test28-without-std-fns-ref.xml:
Likewise.
* tests/data/test-read-write/test28-without-std-vars-ref.xml:
Likewise.
* tests/data/test-read-write/test6.xml: Likewise.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
While looking at the remaining potential sources of instability in the
generation of type ids at abixml generation time, it occurred to me
that we still had several non canonicalized types in the system.
When a given type T is not canonicalized,
hash_as_canonical_type_or_constant later hashes it as an arbitrary
constant. So in the ID hash map of the abixml writer, all the
non-canonicalized types will tend to end-up in the same hash map
bucket, so to speak. More precisely, if T is equivalent to another
type T' which is canonicalized, then T and T' will end-up in different
buckets (in the same hash map) because they will have different hash
values as returned by hash_as_canonical_type_or_constant. They will
thus end-up having different type-ids because they are in different
buckets.
The solution is to make sure that T is *also* canonicalized. That
way, if T and T' are equivalent, they'll end-up in the same bucket and
they will have the same type-id.
In other words, all types should be canonicalized now.
The only exception to that rule is declaration-only classes and
unions. The reason why a declaration-only type 'struct foo' needs to
stay non-canonicalized is that it must equal all the definitions of
'struct foo' that can be found elsewhere.
This patch thus canonicalizes all types that were not still not being
canonicalized. It also adds an assert in
hash_as_canonical_type_or_constant to ensure that only
declaration-only class_or_union types are not canonicalized.
* include/abg-fwd.h (is_declaration_only_class_or_union_type):
Declare new ...
* src/abg-ir.cc (is_declaration_only_class_or_union_type):
... function.
(clone_array): Add the cloned array subrange to
its scope so that it can later be canonicalized.
(synthesize_type_from_translation_unit)
(synthesize_function_type_from_translation_unit): Canonicalize the
synthesized types.
(hash_as_canonical_type_or_constant): Ensure that all types are
canonicalized.
* src/abg-dwarf-reader.cc (maybe_canonicalize_type): Remove
useless overload.
(build_ir_node_for_variadic_parameter_type)
(schedule_array_tree_for_late_canonicalization): Define new static
functions.
(maybe_strip_qualification): Schedule type canonicalization for
types cloned prior to editing.
(build_function_type): Use the new
build_ir_node_for_variadic_parameter_type. It takes care of
canonicalizing variadic parameter types.
(build_function_decl): Canonicalize the function type that is
created here.
(build_ir_node_from_die): Use the overload of
maybe_canonicalize_type which canonicalizes class_or_union nodes
directly, rather than the one which handles DIE offsets. The
latter was used as an optimization to reduce the size of the array
of types scheduled for canonicalization, as DIE offsets take less
space than pointers to IR types. Now that we have DIE
de-duplication, my bet is that we can do away with the former.
And that also ensures that we miss no type for canonicalization
purposes.
* src/abg-reader.cc (build_array_type_def): Canonicalize the
subrange types of the array.
(build_type): Canonicalize all types.
* tests/data/test-annotate/libtest23.so.abi: Adjust.
* tests/data/test-annotate/libtest24-drop-fns-2.so.abi: Likewise.
* tests/data/test-annotate/libtest24-drop-fns.so.abi: Likewise.
* tests/data/test-annotate/test0.abi: Likewise.
* tests/data/test-annotate/test13-pr18894.so.abi: Likewise.
* tests/data/test-annotate/test15-pr18892.so.abi: Likewise.
* tests/data/test-annotate/test17-pr19027.so.abi: Likewise.
* tests/data/test-annotate/test18-pr19037-libvtkRenderingLIC-6.1.so.abi:
Likewise.
* tests/data/test-annotate/test19-pr19023-libtcmalloc_and_profiler.so.abi:
Likewise.
* tests/data/test-annotate/test20-pr19025-libvtkParallelCore-6.1.so.abi:
Likewise.
* tests/data/test-annotate/test21-pr19092.so.abi: Likewise.
* tests/data/test-diff-filter/test30-pr18904-rvalueref-report0.txt:
Likewise.
* tests/data/test-diff-filter/test30-pr18904-rvalueref-report1.txt:
Likewise.
* tests/data/test-diff-filter/test30-pr18904-rvalueref-report2.txt:
Likewise.
* tests/data/test-diff-filter/test35-pr18754-no-added-syms-report-0.txt:
Likewise.
* tests/data/test-read-dwarf/PR22015-libboost_iostreams.so.abi:
Likewise.
* tests/data/test-read-dwarf/PR22122-libftdc.so.abi: Likewise.
* tests/data/test-read-dwarf/PR25042-libgdbm-clang-dwarf5.so.6.0.0.abi:
Likewise.
* tests/data/test-read-dwarf/libtest23.so.abi: Likewise.
* tests/data/test-read-dwarf/libtest24-drop-fns-2.so.abi:
Likewise.
* tests/data/test-read-dwarf/libtest24-drop-fns.so.abi: Likewise.
* tests/data/test-read-dwarf/test-libandroid.so.abi: Likewise.
* tests/data/test-read-dwarf/test0.abi: Likewise.
* tests/data/test-read-dwarf/test0.hash.abi: Likewise.
* tests/data/test-read-dwarf/test10-pr18818-gcc.so.abi: Likewise.
* tests/data/test-read-dwarf/test11-pr18828.so.abi: Likewise.
* tests/data/test-read-dwarf/test12-pr18844.so.abi: Likewise.
* tests/data/test-read-dwarf/test13-pr18894.so.abi: Likewise.
* tests/data/test-read-dwarf/test15-pr18892.so.abi: Likewise.
* tests/data/test-read-dwarf/test16-pr18904.so.abi: Likewise.
* tests/data/test-read-dwarf/test17-pr19027.so.abi: Likewise.
* tests/data/test-read-dwarf/test18-pr19037-libvtkRenderingLIC-6.1.so.abi:
Likewise.
* tests/data/test-read-dwarf/test19-pr19023-libtcmalloc_and_profiler.so.abi:
Likewise.
* tests/data/test-read-dwarf/test20-pr19025-libvtkParallelCore-6.1.so.abi:
Likewise.
* tests/data/test-read-dwarf/test21-pr19092.so.abi: Likewise.
* tests/data/test-read-dwarf/test22-pr19097-libstdc++.so.6.0.17.so.abi:
Likewise.
* tests/data/test-read-dwarf/test9-pr18818-clang.so.abi: Likewise.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
When emitting the declarations of a given translation unit, those
declarations are not sorted. Ooops. This patch adds topological
sorting for those declarations, making the decls defined first to be
emitted first. When the decls are defined at the same location then
the pretty representation is used for lexicographic sorting instead.
It turns out that during the topological sorting for types there was
some uncertainty when the declarations of the types had the same
definition location. This patch re-uses the declaration sorting above
for the declarations of these types.
* include/abg-ir.h (scope_decl::get_sorted_member_decls): Declare
new member function.
* src/abg-ir.cc (struct decl_topo_comp): New sorting functor.
(type_topo_comp::operator()): Re-use the decl_topo_comp to sort
type declarations.
(scope_decl::priv::sorted_members_): Add new data member.
(scope_decl::get_sorted_member_decls): Define new member function.
* src/abg-writer.cc (write_translation_unit): Use the new
scope_decl::get_sorted_member_decls.
* tests/data/test-annotate/libtest23.so.abi: Adjust.
* tests/data/test-annotate/test15-pr18892.so.abi: Likewise.
* tests/data/test-annotate/test17-pr19027.so.abi: Likewise.
* tests/data/test-annotate/test18-pr19037-libvtkRenderingLIC-6.1.so.abi:
Likewise.
* tests/data/test-annotate/test19-pr19023-libtcmalloc_and_profiler.so.abi:
Likewise.
* tests/data/test-annotate/test20-pr19025-libvtkParallelCore-6.1.so.abi:
Likewise.
* tests/data/test-annotate/test21-pr19092.so.abi: Likewise.
* tests/data/test-read-dwarf/libtest23.so.abi: Likewise.
* tests/data/test-read-dwarf/test-libandroid.so.abi: Likewise.
* tests/data/test-read-dwarf/test15-pr18892.so.abi: Likewise.
* tests/data/test-read-dwarf/test17-pr19027.so.abi: Likewise.
* tests/data/test-read-dwarf/test18-pr19037-libvtkRenderingLIC-6.1.so.abi:
Likewise.
* tests/data/test-read-dwarf/test19-pr19023-libtcmalloc_and_profiler.so.abi:
Likewise.
* tests/data/test-read-dwarf/test20-pr19025-libvtkParallelCore-6.1.so.abi:
Likewise.
* tests/data/test-read-dwarf/test21-pr19092.so.abi: Likewise.
* tests/data/test-read-dwarf/test22-pr19097-libstdc++.so.6.0.17.so.abi:
Likewise.
* tests/data/test-read-write/test2.xml: Likewise.
* tests/data/test-read-write/test28-without-std-fns-ref.xml:
Likewise.
* tests/data/test-read-write/test28-without-std-vars-ref.xml:
Likewise.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
CV-qualifiers of a typedef of an array type apply to the elements of
the array. So this can be transformed into a typedef of array type
which element type is similarly CV-qualified.
That transformation helps avoiding spurious changes that might occur
when comparing the latter form against the former even though both are
equivalent.
This patch performs that transformation, just like we already do for
CV-qualified array types which are transformed into an array of
similarly CV-qualified elements.
Performing that transformation amounts to editing the type of the
array elements. As those types might be used by other parts of the
type graph (type node sharing), the patch "clones" the type sub-tree
of interest and edits the cloned version. That way, the shared type
nodes are not edited. It appears that in the existing version of
maybe_strip_qualification, the transformation of CV-qualified arrays
into arrays of CV-qualified elements was unfortunately editing shared
type nodes. The patch fixes that and re-works the logic of*
maybe_strip_qualification to make it handle this new case and the
previous one in a somewhat generic manner.
* include/abg-fwd.h (is_typedef_of_array, clone_array)
(clone_typedef, clone_qualified_type, clone_array_tree): Declare
new functions.
(peel_qualified_or_typedef_type): Declare new overload.
(is_array_of_qualified_element): Constify the parameter.
* include/abg-ir.h ({qualified_type,
typedef}_def::set_underlying_type): Add new member functions.
(array_type_def::subrange_type::subrange_type): Make constify the
reference to the underlying type parameter.
* src/abg-ir.cc (is_array_of_qualified_element): Constify the
parameter.
(peel_qualified_or_typedef_type): Define new
overload for type_base_sptr.
(clone_typedef_array_qualified_type): Define static function.
(clone_array clone_typedef, clone_qualified_type)
(clone_array_tree, is_typedef_of_array): Define new functions.
(qualified_type_def::get_underlying_type): Rename the return type
shared_ptr<type_base> into type_base_sptr.
({typedef, qualified_type}_def::set_underlying_type): Define new
member function.
(array_type_def::subrange_type::priv::priv): Initialize the
'infinite_' data member.
* src/abg-dwarf-reader.cc (maybe_strip_qualification): Handle
qualified typedef of arrays. Merge this with the handling of
qualified arrays. Note that before editing the elements of the
array to make the array (or typedef) qualifier apply to the
element the sub-tree is cloned to make its type nodes be
'un-shared'. This prevents us from editing type nodes that are
shared by other type expressions.
* tests/data/test-diff-filter/test-PR26739-report-0.txt: New
reference test output.
* tests/data/test-diff-filter/test-PR26739-2-report-0.txt: Likewise.
* tests/data/test-diff-filter/test-PR26739-v{0,1}.c: Source code
of new binary test input.
* tests/data/test-diff-filter/test-PR26739-2-v{0,1}.c: Likewise.
* tests/data/test-diff-filter/test-PR26739-v{0,1}.o: New binary
test inputs.
* tests/data/test-diff-filter/test-PR26739-2-v{0,1}.o: Likewise.
* tests/data/Makefile.am: Add the new test material above to
source distribution.
* tests/test-diff-filter.cc (in_out_specs): Add the test inputs
above to this harness.
* tests/data/test-annotate/test15-pr18892.so.abi: Adjust.
* tests/data/test-annotate/test17-pr19027.so.abi: Likewise.
* tests/data/test-annotate/test19-pr19023-libtcmalloc_and_profiler.so.abi:
Likewise.
* tests/data/test-annotate/test21-pr19092.so.abi: Likewise.
* tests/data/test-read-dwarf/PR22122-libftdc.so.abi: Likewise.
* tests/data/test-read-dwarf/test12-pr18844.so.abi: Likewise.
* tests/data/test-read-dwarf/test15-pr18892.so.abi: Likewise.
* tests/data/test-read-dwarf/test17-pr19027.so.abi: Likewise.
* tests/data/test-read-dwarf/test19-pr19023-libtcmalloc_and_profiler.so.abi: Likewise.
* tests/data/test-read-dwarf/test21-pr19092.so.abi: Likewise.
* tests/data/test-read-dwarf/test22-pr19097-libstdc++.so.6.0.17.so.abi: Likewise.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
Update this test reference output as per the commit below:
b7ad7f2b Bug 26770 - Spurious declaration-only-ness induces spurious type changes
* tests/data/test-read-dwarf/test-libandroid.so.abi: Update.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
Location expressions might occasionally be of length 0. E.g. a reason
for that are thread local variables that do not exactly have a
location to refer to. Compilers/Linkers may choose an empty location
description. E.g. see the dwarfdump output for the added testcase
based on libandroid.so (from AOSP).
$ dwarfdump libandroid.so|egrep -B9 "gChoreographerE$"
LOCAL_SYMBOLS:
< 1><0x00000022> DW_TAG_namespace
DW_AT_name android
< 2><0x00000027> DW_TAG_variable
DW_AT_name gChoreographer
DW_AT_type <0x00000b65>
DW_AT_decl_file 0x00000003 .../choreographer.cpp
DW_AT_decl_line 0x00000059
DW_AT_location len 0x0000: :
DW_AT_linkage_name _ZN7androidL14gChoreographerE
The DW_AT_location is properly read by elfutils' dwarf_location(), but
is not useful for us to proceed with. Hence early exit on this.
* src/abg-dwarf-reader.cc (die_location_expr): Ignore zero
length location expressions.
* tests/data/Makefile.am: Add new test files.
* tests/data/test-read-dwarf/test-libandroid.so: New test file.
* tests/data/test-read-dwarf/test-libandroid.so.abi: Likewise.
* tests/test-read-dwarf.cc: Add new test case.
Reported-by: Dan Albert <danalbert@google.com>
Reviewed-by: Giuliano Procida <gprocida@google.com>
Cc: Mark Wielaard <mark@klomp.org>
Signed-off-by: Matthias Maennich <maennich@google.com>
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
The ordering of canonical types (in an abi-instr XML element) appears
to be sensitive to the particular C++ library used and the presence of
other threads doing heap allocation. This patch forces distinct
synthetic enum-underlying types to have distinct names, which ensures
deterministic XML output order.
* src/abg-dwarf-reader.cc
(build_internal_underlying_enum_type_name): Add a size
argument (and don't default is_anonymous argument). Append
size of type to synthetic type name.
(build_enum_underlying_type): Pass type size to
build_internal_underlying_enum_type_name.
* tests/data/test-abidiff-exit/test-decl-enum-report-3.txt:
Update. Note that there may be an issue with leaf-mode
reporting of pointer type changes.
* tests/data/test-diff-dwarf-abixml/PR25409-librte_bus_dpaa.so.20.0.abi:
Regenerate this (catching up with various abidw updates).
* tests/data/test-annotate/test-anonymous-members-0.o.abi:
Refresh with new type names.
* tests/data/test-annotate/test0.abi: Likewise.
* tests/data/test-annotate/test13-pr18894.so.abi: Likewise.
* tests/data/test-annotate/test14-pr18893.so.abi: Likewise.
* tests/data/test-annotate/test15-pr18892.so.abi: Likewise.
* tests/data/test-annotate/test17-pr19027.so.abi: Likewise.
* tests/data/test-annotate/test18-pr19037-libvtkRenderingLIC-6.1.so.abi:
Likewise.
* tests/data/test-annotate/test19-pr19023-libtcmalloc_and_profiler.so.abi:
Likewise.
* tests/data/test-annotate/test20-pr19025-libvtkParallelCore-6.1.so.abi:
Likewise.
* tests/data/test-annotate/test21-pr19092.so.abi: Likewise.
* tests/data/test-read-dwarf/PR22015-libboost_iostreams.so.abi:
Likewise.
* tests/data/test-read-dwarf/PR22122-libftdc.so.abi: Likewise.
* tests/data/test-read-dwarf/PR25007-sdhci.ko.abi: Likewise.
* tests/data/test-read-dwarf/PR25042-libgdbm-clang-dwarf5.so.6.0.0.abi:
Likewise.
* tests/data/test-read-dwarf/test0.abi: Likewise.
* tests/data/test-read-dwarf/test0.hash.abi: Likewise.
* tests/data/test-read-dwarf/test10-pr18818-gcc.so.abi: Likewise.
* tests/data/test-read-dwarf/test11-pr18828.so.abi: Likewise.
* tests/data/test-read-dwarf/test12-pr18844.so.abi: Likewise.
* tests/data/test-read-dwarf/test13-pr18894.so.abi: Likewise.
* tests/data/test-read-dwarf/test14-pr18893.so.abi: Likewise.
* tests/data/test-read-dwarf/test15-pr18892.so.abi: Likewise.
* tests/data/test-read-dwarf/test16-pr18904.so.abi: Likewise.
* tests/data/test-read-dwarf/test17-pr19027.so.abi: Likewise.
* tests/data/test-read-dwarf/test18-pr19037-libvtkRenderingLIC-6.1.so.abi:
Likewise.
* tests/data/test-read-dwarf/test19-pr19023-libtcmalloc_and_profiler.so.abi:
Likewise.
* tests/data/test-read-dwarf/test20-pr19025-libvtkParallelCore-6.1.so.abi:
Likewise.
* tests/data/test-read-dwarf/test21-pr19092.so.abi: Likewise.
* tests/data/test-read-dwarf/test22-pr19097-libstdc++.so.6.0.17.so.abi:
Likewise.
* tests/data/test-read-dwarf/test9-pr18818-clang.so.abi: Likewise.
Signed-off-by: Giuliano Procida <gprocida@google.com>
Signed-off-by: Matthias Maennich <maennich@google.com>
The functor virtual_member_function_less_than did not take into
account linkage name which can be the only difference when multiple
destructors with differing mangled names are present.
This change adds a check for linkage names and also flattens the
control flow in the comparison method to make the logic clearer.
Lastly, this change also uses std::stable_sort, in case all that
remains is insertion order.
* src/abg-ir.cc
(virtual_member_function_less_than::operator()): Name
temporaries like offsets and symbols to reduce repetition;
test each pair of elements (including symbol presence) and
return immediately if there's a difference; add a comparison
of linkage name just after comparing symbol names.
(sort_virtual_member_functions): Use stable_sort instead of
sort.
* tests/data/test-read-dwarf/PR22015-libboost_iostreams.so.abi:
Update with new ordering of member functions.
* tests/data/test-read-dwarf/test22-pr19097-libstdc++.so.6.0.17.so.abi:
Likewise.
Signed-off-by: Giuliano Procida <gprocida@google.com>
Signed-off-by: Matthias Maennich <maennich@google.com>
Sometimes when amending a C++ class to add new members or properties
to it as directed by the DWARF debug information, we can end-up with
discrepancies related to declaration-only-ness. That is, an instances
of a given type Foo can be wrongly assigned declaration-only-ness that
should have been only carried by another instance Foo. Then, later,
comparing two pointers to Foo might wrongly lead to spurious reported
changes due to the spurious differences of declaration-only-ness in
two instances of Foo.
By fixing the setting of the declaration-only-ness, especially when
amending a C++ class this patch fixes that spurious change detected.
* src/abg-dwarf-reader.cc (add_or_update_class_type): When
creating a class, set declaration-only-ness unconditionally. When
updating the class however, only set the declaration-only-ness
when the current one is not consistent with the size of the class.
* tests/data/test-annotate/test14-pr18893.so.abi: Adjust.
* tests/data/test-annotate/test19-pr19023-libtcmalloc_and_profiler.so.abi: Likewise.
* tests/data/test-annotate/test20-pr19025-libvtkParallelCore-6.1.so.abi: Likewise.
* tests/data/test-read-dwarf/test14-pr18893.so.abi: Likewise.
* tests/data/test-read-dwarf/test19-pr19023-libtcmalloc_and_profiler.so.abi: Likewise.
* tests/data/test-read-dwarf/test20-pr19025-libvtkParallelCore-6.1.so.abi: Likewise.
* tests/data/test-read-dwarf/test22-pr19097-libstdc++.so.6.0.17.so.abi: Likewise.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
When using relative compilation unit paths in DWARF, the lookup for an
existing one was done with an incorrect path. In particular, a '/' was
prefixed to the path regardless of whether the compilation_dir is set.
That led to the instantiation of an additional translation unit that and
failed when adding it to the corpus due to violating translation unit
uniqueness. Fix that by correcting the lookup.
* src/abg-dwarf-reader.cc(build_translation_unit_and_add_to_ir):
Fix lookup for potentially already existing translation units.
Reported-by: Dan Albert <danalbert@google.com>
Signed-off-by: Matthias Maennich <maennich@google.com>
Some types like unnamed-enum-underlying-type are not distinguished by
type_topo_comp. This can result in nondeterministic output and flakey
tests.
While a more complete ordering from type_topo_comp would be nice, the
nondeterminism can reduced by preserving the relative order of
identically-named types.
* src/abg-ir.cc (scope_decl::get_sorted_canonical_types): Sort
canonical types with std::stable_sort(..., type_topo_comp()).
Signed-off-by: Giuliano Procida <gprocida@google.com>
Reviewed-by: Matthias Maennich <maennich@google.com>
Signed-off-by: Matthias Maennich <maennich@google.com>
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
When running abidiff on test34-libjemalloc.so.2-intel-16.0.3 it would
crash in array_type_def::subrange_type::get_length on the ABG_ASSERT
get_upper_bound() >= get_lower_bound(). This was because that file
contained a subrange upper_bound value encoded with data1 or data2
without an underlying type. In that case we assumed the value was
encoded as a signed value which caused some of the upper bounds to
be negative (while the lower bound, which wasn't given, was assumed
to be zero).
* src/abg-dwarf-reader.cc (build_subrange_type): Default
is_signed to false.
Signed-off-by: Mark Wielaard <mark@klomp.org>
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
This patch was originally submitted by in a comment of a problem
report at https://sourceware.org/bugzilla/show_bug.cgi?id=26684#c10.
When reading the bounds of a subrange_type, we need to know if the
constant value of those bounds is signed or not.
For that, we used to just look at the form of those constant and infer
the signedness from there.
The problem is that the signedness is really determined by the value
of the DW_AT_encoding attribute present on the underlying type of the
subrange_type; the form of the bound is mere implementation detail
that is downstream of the information of carried by the DW_AT_encoding
attribute.
This patch thus does the right thing by looking at the underlying type
of the subrange_type and by doing away with the unnecessary messing
with attribute value forms.
* src/abg-dwarf-reader.cc (die_attribute_has_form)
(die_attribute_is_signed, die_attribute_is_unsigned)
(die_attribute_has_no_signedness): Remove static functions.
(die_constant_attribute): Add the 'is_signed' parameter.
(die_address_attribute): Adjust comment.
(build_subrange_type): Determine signedness of the bounds by
looking at the DW_AT_encoding attribute of the underlying type.
Signed-off-by: Mark Wielaard <mark@klomp.org>
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
When working in development environments with compiler versions that
might be very bleeding edge (like the Fedora Rawhide distribution) it
might be worthwhile to disable all compiler optimization to have a
better debugging experience. In practice, I bumped into this need
again and again.
So I am adding this ABIGAIL_NO_OPTIMIZATION_DEBUG environment variable
to basically allow the "-g -O0" combination, if need be.
This patch obviously doesn't change any existing behaviour if the user
doesn't set this newly introduced environment variable.
* configure.ac: Set the CXXFLAGS and CFLAGS to "-g -O0 -Wall
-Wextra -Werror" if the ABIGAIL_NO_OPTIMIZATION_DEBUG is set.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
In the abixml writer, we recently stopped using
type_base::dynamic_hash as a slow path to hash a non-canonicalized
type. Rather, we use an arbitrary constant as a hash for those
non-canonicalized types. This amounts to using structural comparison
for those types. The function that implements this hashing scheme is
hash_as_canonical_type_or_constant.
A potential concern for that approach was the possible negative impact
on speed. As it turns out since the change went in, there was no
noticeable speed impact raised after testing.
In insight, this is understandable as the number of non-canonicalized
types should be extremely reduced now that we canonicalize pretty much
all types. As far as I understand it at this point, the only types
that would not be canonicalized are unresolved declaration-only
types. And, all in all, structurally comparing unresolved
declaration-only types should be "fast enough".
If we ever stumble across any other non-canonicalized type, I think
that would be an artifact of a bug that ought to be fixed.
On that basis, I went ahead and used
hash_as_canonical_type_or_constant throughout the code base and did
away with the use of type_base::dynamic_hash for now, until it's
properly audited, regression tested and got ready for the use cases
where it might make sense.
This patch thus makes hash_type use
hash_as_canonical_type_or_constant. The writer is then back to using
hash_type again, as it used to; but at the same time, it's still using
structural comparison for non-canonilized types. So is
hash_type_or_decl, which now uses hash_type to hash types, rather than
using type_base::dynamic_hash. Note that the comparison engine
heavily uses hash_type_or_decl to hash diff nodes.
So with this small patch the comparison engine is now using structural
comparison of non-canonicalized types (and diff nodes), just as the
abixml writer does.
* include/abg-fwd.h (hash_as_canonical_type_or_constant): Remove
public declaration of this function.
* src/abg-hash.cc (type_base::dynamic_hash::operator()): Add a
comment.
* src/abg-ir.cc (hash_as_canonical_type_or_constant): Make this
function static now.
(hash_type_or_decl): Use hash_type for types.
* src/abg-writer.cc (type_hasher::operator()): Use hash_type.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
A recent change[1] triggered a variation in the ABI changes reported,
depending on the platform.
It turned out the change made the redundancy detector walk more diff
nodes (like inserted/deleted virtual member functions and their
implicit parameter type) and that uncovered several underlying issues
that has been latent for a long time.
First, we were not walking the inserted/deleted virtual member
functions in a deterministic manner at reporting time. Rather than
walking the unordered maps containing those functions, this patch now
walk them in lexicographic order. The patch also does something
similar for the changed data members, but this time during the diff
graph analysis. That order affects how we consider a given type
change to be redundant.
Second, when looking a diff node named N, if another diff node N'
equivalent to N has already been marked redundant (and thus filtered
out already), we were sometimes wrongly failing to detect and mark N
as redundant. This patch fixes that.
I realized that some code was now unnecessary so I removed it.
A lot of reference output of tests are adjusted by this patch.
Mostly, these were cases we were failing to properly detect (and
filter out) as redundant reports. So the change reports should
hopefully look more concise and to the point now.
[1] the recent change is this one:
2f92777d Consider the implicit 'this' parameter when comparing methods
* src/abg-comparison-priv.h
(diff_context::priv::last_visited_diff_node_): Remove unnecessary
data member.
(class_or_union_diff::priv::sorted_{deleted,inserted}_member_functions_):
Add new data members.
(sort_string_member_function_sptr_map): Declare new function.
* src/abg-comparison.cc (sort_string_member_function_sptr_map):
Define new function.
(redundancy_marking_visitor::visit_begin): If the current diff
node is equivalent to another one that has been already marked
redundant, then consider the current diff node as redundant as
well. Considering the fact an ancestor node has been filtered out
is now useless because if that's the case then the current
descendant node wouldn't even be walked at reporting time. So
remove the call to diff_has_ancestor_filtered_out.
(categorize_redundancy): Remove useless call here as well.
(diff_has_ancestor_filtered_out, diff_has_ancestor_filtered_out)
(diff_context::{mark_last_diff_visited_per_class_of_equivalence,
clear_last_diffs_visited_per_class_of_equivalence,
get_last_visited_diff_of_class_of_equivalence}): Remove
unnecessary functions.
(redundancy_marking_visitor::visit_end): Add comment.
(class_diff::ensure_lookup_tables_populated): Lexicographically
sort inserted/deleted member functions.
(class_or_union_diff::chain_into_hierarchy): Chain changed data
members diff nodes in a sorted manner.
* src/abg-default-reporter.cc (default_reporter::report): Report
deleted/inserted member functions in lexicographic order.
* tests/data/test-diff-filter/test31-pr18535-libstdc++-report-0.txt:
Adjust.
* tests/data/test-diff-filter/test31-pr18535-libstdc++-report-1.txt:
Likewise.
* tests/data/test-diff-pkg/libICE-1.0.6-1.el6.x86_64.rpm--libICE-1.0.9-2.el7.x86_64.rpm-report-0.txt:
Likewise.
* tests/data/test-diff-pkg/tbb-4.1-9.20130314.fc22.x86_64--tbb-4.3-3.20141204.fc23.x86_64-report-0.txt:
Likewise.
* tests/data/test-diff-pkg/tbb-4.1-9.20130314.fc22.x86_64--tbb-4.3-3.20141204.fc23.x86_64-report-1.txt:
Likewise.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
Since 2013 the implicit 'this' parameter has been excluded from the
function parameters taken into account while comparing class member
functions. This was an early measure to avoid infinite recursion that
would then occur when comparing classes (and thus their member
functions that are referenced in their vtable). But since then, we've
built descent infrastructure to prevent this kind of recursion in a
more generic manner.
This patch thus removes that restriction and should therefore lift the
concerns expressed in the bug
https://sourceware.org/bugzilla/show_bug.cgi?id=26672.
Namely, changes to (data members of) a class should now be detected
when comparing member functions of that class.
With this change, the reference output of several comparison
regression tests changed because, obviously, some impacted member
functions are now reported along with detecting changes in data
membrers of classes. The patch thus adjusts those reference ouputs.
The patch also adjust the behaviour of the predicate:
"accessed_through = pointer|reference|reference-or-pointer"
The idea is to make the predicate work on qualified version of a type.
* include/abg-ir.h (function_type::get_first_parm): Declare new
accessor.
* src/abg-ir.cc (function_type::get_first_parm): Define new
accessor.
(equals): In the overload for function_type,
always take the implicit "this" parameter into account in
parameter comparisons.
(function_type::get_first_non_implicit_parm): Adjust comment.
* src/abg-comp-filter.cc (function_name_changed_but_not_symbol):
Avoid potential NULL pointer dereferencing.
* src/abg-comparison.cc
(function_type_diff::ensure_lookup_tables_populated): Always take
the changes to the implicit 'this' parameter into account in the
function type diff.
(compute_diff): In the overload for function_type, Always compare
the implicit 'this' parameter when comparing function parameters.
* src/abg-default-reporter.cc (default_reporter::report): Refer to
"implicit parameter" when reporting changes on parameters
artificially generated by the compiler.
* src/abg-suppression.cc (type_suppression::suppresses_diff): Make
the 'access_through' predicate work on a qualified version of type
'S', even if it was meant to work on type 'S'. This allows it to
work on 'const S', especially when S is accessed through 'pointer
to const S', which happens when we consider the implicit 'this'
parameter of a const member function.
* tests/data/test-abicompat/test5-fn-changed-report-0.txt: Adjust.
* tests/data/test-abicompat/test5-fn-changed-report-1.txt: Likewise.
* tests/data/test-abidiff-exit/test1-voffset-change-report0.txt:
Likewise.
* tests/data/test-abidiff/test-PR18791-report0.txt: Likewise.
* tests/data/test-abidiff/test-struct1-report.txt: Likewise.
* tests/data/test-diff-dwarf-abixml/test0-pr19026-libvtkIOSQL-6.1.so.1-report-0.txt:
Likewise.
* tests/data/test-diff-dwarf/test0-report.txt: Likewise.
* tests/data/test-diff-dwarf/test28-vtable-changes-report-0.txt: Likewise.
* tests/data/test-diff-dwarf/test29-vtable-changes-report-0.txt: Likewise.
* tests/data/test-diff-dwarf/test30-vtable-changes-report-0.txt: Likewise.
* tests/data/test-diff-dwarf/test31-vtable-changes-report-0.txt: Likewise.
* tests/data/test-diff-dwarf/test36-ppc64-aliases-report-0.txt: Likewise.
* tests/data/test-diff-dwarf/test41-PR20476-hidden-report-0.txt: Likewise.
* tests/data/test-diff-dwarf/test42-PR21296-clanggcc-report0.txt: Likewise.
* tests/data/test-diff-dwarf/test5-report.txt: Likewise.
* tests/data/test-diff-dwarf/test8-report.txt: Likewise.
* tests/data/test-diff-filter/test0-report.txt: Likewise.
* tests/data/test-diff-filter/test01-report.txt: Likewise.
* tests/data/test-diff-filter/test10-report.txt: Likewise.
* tests/data/test-diff-filter/test13-report.txt: Likewise.
* tests/data/test-diff-filter/test2-report.txt: Likewise.
* tests/data/test-diff-filter/test28-redundant-and-filtered-children-nodes-report-0.txt:
Likewise.
* tests/data/test-diff-filter/test28-redundant-and-filtered-children-nodes-report-1.txt:
Likewise.
* tests/data/test-diff-filter/test30-pr18904-rvalueref-report0.txt:
Likewise.
* tests/data/test-diff-filter/test30-pr18904-rvalueref-report1.txt:
Likewise.
* tests/data/test-diff-filter/test30-pr18904-rvalueref-report2.txt:
Likewise.
* tests/data/test-diff-filter/test31-pr18535-libstdc++-report-0.txt:
Likewise.
* tests/data/test-diff-filter/test31-pr18535-libstdc++-report-1.txt:
Likewise.
* tests/data/test-diff-filter/test35-pr18754-no-added-syms-report-0.txt:
Likewise.
* tests/data/test-diff-filter/test35-pr18754-no-added-syms-report-1.txt:
Likewise.
* tests/data/test-diff-filter/test4-report.txt: Likewise.
* tests/data/test-diff-filter/test41-report-0.txt: Likewise.
* tests/data/test-diff-filter/test9-report.txt: Likewise.
* tests/data/test-diff-pkg/libsigc++-2.0-0c2a_2.4.0-1_amd64--libsigc++-2.0-0v5_2.4.1-1ubuntu2_amd64-report-0.txt:
Likewise.
* tests/data/test-diff-pkg/tbb-4.1-9.20130314.fc22.x86_64--tbb-4.3-3.20141204.fc23.x86_64-report-0.txt:
Likewise.
* tests/data/test-diff-pkg/tbb-4.1-9.20130314.fc22.x86_64--tbb-4.3-3.20141204.fc23.x86_64-report-1.txt:
Likewise.
* tests/data/test-diff-suppr/test24-soname-report-0.txt: Likewise.
* tests/data/test-diff-suppr/test24-soname-report-1.txt: Likewise.
* tests/data/test-diff-suppr/test24-soname-report-10.txt: Likewise.
* tests/data/test-diff-suppr/test24-soname-report-11.txt: Likewise.
* tests/data/test-diff-suppr/test24-soname-report-12.txt: Likewise.
* tests/data/test-diff-suppr/test24-soname-report-13.txt: Likewise.
* tests/data/test-diff-suppr/test24-soname-report-14.txt: Likewise.
* tests/data/test-diff-suppr/test24-soname-report-15.txt: Likewise.
* tests/data/test-diff-suppr/test24-soname-report-16.txt: Likewise.
* tests/data/test-diff-suppr/test24-soname-report-2.txt: Likewise.
* tests/data/test-diff-suppr/test24-soname-report-3.txt: Likewise.
* tests/data/test-diff-suppr/test24-soname-report-4.txt: Likewise.
* tests/data/test-diff-suppr/test24-soname-report-5.txt: Likewise.
* tests/data/test-diff-suppr/test24-soname-report-6.txt: Likewise.
* tests/data/test-diff-suppr/test24-soname-report-7.txt: Likewise.
* tests/data/test-diff-suppr/test24-soname-report-8.txt: Likewise.
* tests/data/test-diff-suppr/test24-soname-report-9.txt: Likewise.
* tests/data/test-diff-suppr/test31-report-1.txt: Likewise.
* tests/data/test-diff-suppr/test33-report-0.txt: Likewise.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
The suppression specification in test38-char-class-in-ini.abignore was
introduced in commit 1478d9cc1c.
Unfortunately it contains two errors. One causes the file name not to
match as the string is the full path, not the base name. The other is
a typo that causes the file name match not to even be attempted. The
two mistakes cancel in the test, but result in a suppression
specification that is broader than intended.
* tests/data/test-diff-suppr/test38-char-class-in-ini.abignore:
Don't anchor regex match to beginning of file name.
Change "filename_regexp" to "file_name_regexp".
Signed-off-by: Giuliano Procida <gprocida@google.com>
If there ever is a discrepancy between the architectures of the
corpuses of a corpus group, libabigail will just abort with an
assertion, if enabled. However, this is a case of invalid input and
the cause should be reported to the user.
* src/abg-corpus.cc (corpus_group::add_corpus): Report
architecture discrepancies.
Signed-off-by: Giuliano Procida <gprocida@google.com>
When building a union type we try to ensure that each member is
present only once. This is because the code to build the union is
also used to actually update a partially constructed union. To do so,
before adding a member to the union, the member is looked up (among
the current members) by name to see if it's already present or not.
But then for anonymous members, the name of the member is empty.
After the first anonymous member is added to the union, subsequent
look-ups with an empty name all succeed. So no more than one
anonymous member is added to the union. Oops.
A way to fix this is to perform the lookup by taking into account the
type of the anonymous data member, not its (empty) name. We already
do this for anonymous data members of classes/structs.
This patch thus uses that type-based anonymous data member lookup for
unions.
But then now that unions can have several anonymous members, another
issue was uncovered.
Array types whose elements are of anonymous type can be wrongly
considered different because of canonicalization issues.
Let's suppose we have these two arrays whose internal pretty
representation are:
"__anonymous_struct_1__ foo[5]"
and
"__anonymous_struct_2__ foo[5]"
These are arrays of 5 elements of type anonymous struct. Suppose the
anonymous structs "__anonymous_struct_1__" and
"__anonymous_struct_2__" are structurally equivalent. Because the
internal names of these array element types are different, the
internal pretty representations of the arrays are different. And thus
the canonical types of the two arrays are different. And that's
wrong. In this particular case, they should have the same canonical
type and thus be considered equivalent.
This patch thus teaches 'get_type_name' to make the internal type
name of anonymous types of a given kind be the same. Thus, making all
arrays of 5 anonymous struct have the same pretty representation:
"__anonymous_struct__ foo[5]"
This gives the type canonicalizer a chance to detect that those arrays
having the same canonical type.
These two changes while being seemingly unrelated need to be bundled
together to fix a number of issues in the existing test reference
outputs because fixing the first one is needed to uncover the later
issue.
* src/abg-dwarf-reader.cc (add_or_update_union_type): Don't use
the empty name of anonymous members in the lookup to ensure that
all data members are unique. Rather, use the whole anonymous
member itself for the lookup, just like is done to handle
anonymous data member in classes/structs.
* src/abg-reader.cc (build_union_decl): Likewise.
* src/abg-ir.cc (get_generic_anonymous_internal_type_name): Define
new static function.
(get_type_name): For internal purposes, make the type name of all
anonymous types of a given kind to be the same. This allows the
internal representation of anonymous types which are based on type
names to all be the same, so that they can be compared among
themselves during type canonicalization.
* tests/data/test-read-dwarf/test-PR26568-{1,2}.c: Source code of
binary test input.
* tests/data/test-read-dwarf/test-PR26568-{1,2}.o: New binary test input.
* tests/data/test-read-dwarf/test-PR26568-{1,2}.o.abi: New
reference test ouput.
* tests/data/Makefile.am: Add the new test material above to
source distribution.
* tests/test-read-dwarf.cc (in_out_specs): Add the new binary test
input above to this test harness.
* tests/data/test-diff-dwarf-abixml/PR25409-librte_bus_dpaa.so.20.0.abi: Adjust.
* tests/data/test-read-dwarf/PR22122-libftdc.so.abi: Likewise.
* tests/data/test-read-dwarf/PR25007-sdhci.ko.abi: Likewise.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
When handling a binary with abidiff or abidw it can be useful to
provide several different header files directories, for the cases
where the header files of the binary are scathered in several
different directories.
It thus becomes possible to invoke abidiff like this:
abidiff --headers-dir1 first-header-dir1 \
--headers-dir1 second-header-dir1 \
--headers-dir2 first-header-dir2 \
--headers-dir2 second-header-dir2 \
binary1 binary2
This patch adds support for that. It also modifies the
tests/test-abidiff-exit.cc test harness to make it take header
directories. With that modification done, a new test is added in that
harness to exercise this new feature.
This should close the feature request over at
https://sourceware.org/bugzilla/show_bug.cgi?id=26565.
* doc/manuals/abidiff.rst: Update documentation for the
--headers-dir{1,2} options.
* doc/manuals/abidw.rst: Likewise for the --header-dir option.
* include/abg-tools-utils.h (gen_suppr_spec_from_headers): Add new
overload that takes a vector of headers root dirs.
* src/abg-tools-utils.cc (gen_suppr_spec_from_headers_root_dir):
Define new function.
(gen_suppr_spec_from_headers): Define a new overload that takes a
vector of head_root_dir strings; it uses the new
gen_suppr_spec_from_headers function. Use the new overload in the
previous one that takes just one head_root_dir string.
* tools/abidiff.cc (options::headers_dirs{1,2}): Rename
option::headers_dir{1,2} into this one and make it be a vector of
strings rather than just a string.
(parse_command_line): Support several --headers-dir{1,2} on the
command line.
(set_diff_context_from_opts, set_suppressions): Adjust.
* tools/abidw.cc (options::headers_dirs): Renamed
options::headers_dir into this and make it be a vector of strings
rather than just a string.
(parse_command_line): Support several --headers-dir on the command
line.
(set_suppressions): Adjust.
* tests/data/test-abidiff-exit/test-headers-dirs/headers-a/header-a-v{0,1}.h:
Header files of new binary test input.
* tests/data/test-abidiff-exit/test-headers-dirs/headers-b/header-b-v{0,1}.h:
Likewise.
* tests/data/test-abidiff-exit/test-headers-dirs/test-headers-dir-v{0,1}.c:
Source code of new binary test input.
* tests/data/test-abidiff-exit/test-headers-dirs/test-headers-dir-report-{1,2}.txt:
Reference output of new binary test input.
* tests/data/test-abidiff-exit/test-headers-dirs/test-headers-dir-v{0,1}.o:
New binary test input.
* tests/data/Makefile.am: Add the new files above to source
distribution.
* tests/test-abidiff-exit.cc (InOutSpec::in_elfv{0,1}_path): Add
new data members.
(in_out_specs): Adjust the content of this array as its type
changed. Also, add two new entries to run the test over the new
binary test inputs above.
(do_prefix_strings): Define new static function.
(main): Use it the new do_prefix_strings here. Make abidiff
use the --header-dir{1,2} option whenever header directories are
specified in an entry of the in_out_specs array.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
In leaf mode, libabigail fails to report changes to the underlying
type of a typedef.
At its core, this is due to the fact that changes to the underlying
type of a typedef are not considered local. As the leaf reporter only
reports local changes (as opposed to non-local changes which are
changes to sub-types) it doesn't detect those non-local typedef
changes.
To handle this, this patch makes changes to the underlying type of a
typedef be considered as local changes. This is like what we already
do for pointer and qualified types.
Now that we have another set of changes to report in the leaf
reporter, we need to handle how to propagate the category and
redundancy status of those changes. The patch does this too.
Also, just like we do pointer and qualified type changes, the patch
avoids marking the diff node carrying the typedef change as being a
leaf change. That way, only existing leaf changes carrying that
typedef diff node will be reported. For instance, a function whose
parameter has a typedef change will be reported because that change to
the function is considered a leaf change. Otherwise, reporting the
typedef (or the pointer or qualified) type change on its own is not
useful unless it impacts those leaf changes that we deem useful.
The patch adds the example given in problem report to the testsuite.
* src/abg-ir.cc (equals): In the overload for typedef_decls,
report changes to the underlying type as being local of kind
LOCAL_TYPE_CHANGE_KIND.
* src/abg-comparison.cc
(leaf_diff_node_marker_visitor::visit_begin): Do not mark typedef
diff node as leaf node.
(suppression_categorization_visitor::visit_end): Propagate the
'suppressed' category of the underlying type to the parent typedef
unless the later has a local non-type change.
(redundancy_marking_visitor::visit_end): Likewise for the
'redundant' category.
* include/abg-reporter.h (report_non_type_typedef_changes): Rename ...
* src/abg-default-reporter.cc (report_non_type_typedef_changes):
... report_local_typedef_changes into this.
* src/abg-leaf-reporter.cc (leaf_reporter::report): Make the leaf
reporter invoke the reporting method of the default reporter for
typedefs as all typedef changes are now local.
* tests/data/test-diff-filter/test-PR26309-report-0.txt: Add new
test reference output.
* tests/data/test-diff-filter/test-PR26309-v{0,1}.o: Add new test
binary input.
* tests/data/test-diff-filter/test-PR26309-v{0,1}.c: Add source
code for new test binary input.
* tests/data/Makefile.am: Add the new text material above to
source distribution.
* tests/test-diff-filter.cc (in_out_specs): Add the new test input
above to this test harness.
* tests/data/test-diff-pkg/spice-server-0.12.4-19.el7.x86_64-0.12.8-1.el7.x86_64-report-3.txt: Adjust.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
When using type maps the hash used for a given type T is usually the
pointer value of the canonical type of T.
In the rare cases where T doesn't have a canonical type, we were using
a 'dynamic hash' computed by recursively walking the components of T
and progressively building a hash that way.
The dynamic hashing code hasn't been updated in a while and would need
some overhaul to support hashing with schemes like MD5 or maybe sha
even. It might be useful for various use cases that have been
proposed by some users over the years but nobody was motivated enough
to implement it.
In the mean time, rather than trying to come up with a fully beefed up
dynamic hashing code, we'd rather just return a constant number for
non canonicalized types. In practise that amounts to forcing the code
of the maps to always use structural comparison for those non
canonicalized types.
Note that the amount of non-canonicalized types should be fairly
small. For now, the only non-canonicalized types should be
declaration-only types and those are quite fast to compare anyway.
This patch thus introduces a new hashing scheme for maps in the
writer which just uses a numerical constant as the hash for
non-canonicalized types.
* include/abg-fwd.h (hash_as_canonical_type_or_constant): Declare ...
* src/abg-ir.cc (hash_as_canonical_type_or_constant): ... new
function.
* src/abg-writer.cc (type_hasher::operator()): Use the new
hash_as_canonical_type_or_constant.
* tests/data/test-read-dwarf/test16-pr18904.so.abi: Adjust.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
