mirror of
git://sourceware.org/git/libabigail.git
synced 2024-12-18 07:54:36 +00:00
e62901963f
When loading two binaries (e.g, when the library is used by abidiff),
and when the second one does have deep types (e.g, classes with
recursively deep hierarchies) with lots of duplicated types in lots of
translation units, canonicalizing the types of the second binaries can
take a *lot* of time, given the quadratic nature of the structural
type comparisons that take place and the cheer number of those type
comparisons (because of the duplication).
There is already an optimization based on the One Definition Rule in
the canonicalization code. That optimization avoids structural
comparison of types of the same corpus which have the same name. But
then, this optimization only works on types of the first corpus.
As soon as we are loading a second corpus, all types being
canonicalized are coming from a corpus that is different from the
first corpus, by definition. So a structural comparison is taking
place for *all* those types.
The patch extends the existing optimization to make it work on the
second corpus being loaded. Once a type from the second corpus is
canonicalized, the canonical type is cached inside the corpus. Then,
later, when a type with the same name has to be canonicalized, the
system looks inside the cache of that corpus to see if there is a
canonicalized type the same name.
I tested the patch on this command:
abipkgdiff --d1 nss-debuginfo-3.19.1-8.el6_7.i686.rpm \
--d2 nss-debuginfo-3.21.0-0.1.el6_7.i686.rpm \
nss-3.19.1-8.el6_7.i686.rpm \
nss-3.21.0-0.1.el6_7.i686.rpm
I whitnessed a x10 speedup, at least.
On binaries that don't have a lot of duplicated deep types, the patch
doesn't have any noticeable effect. At lesat It doesn't slow things
down in that case.
* include/abg-corpus.h (corpus::{record_canonical_type,
lookup_canonical_type}): Declare new member functions.
* src/abg-corpus.cc (corpus::priv::canonical_types_): New data
member.
(corpus::{record_canonical_type, lookup_canonical_type}): Define
new member functions.
* src/abg-ir.cc (type_base::get_canonical_type_for): Cache the
canonical type inside the corpus of the type being canonicalized.
Then later when canonicalizing another type, lookup in the cache
inside its corpus to see if there is a type with the same name.
* tests/data/test-diff-pkg/tbb-4.1-9.20130314.fc22.x86_64--tbb-4.3-3.20141204.fc23.x86_64-report-0.txt:
Adjust.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
|
||
|---|---|---|
| .. | ||
| data | ||
| Makefile.am | ||
| print-diff-tree.cc | ||
| runtestcanonicalizetypes.sh.in | ||
| test-abicompat.cc | ||
| test-abidiff-exit.cc | ||
| test-abidiff.cc | ||
| test-alt-dwarf-file.cc | ||
| test-core-diff.cc | ||
| test-diff2.cc | ||
| test-diff-dwarf-abixml.cc | ||
| test-diff-dwarf.cc | ||
| test-diff-filter.cc | ||
| test-diff-pkg.cc | ||
| test-diff-suppr.cc | ||
| test-dot.cc | ||
| test-ir-walker.cc | ||
| test-lookup-syms.cc | ||
| test-read-dwarf.cc | ||
| test-read-write.cc | ||
| test-svg.cc | ||
| test-types-stability.cc | ||
| test-utils.cc | ||
| test-utils.h | ||
| test-write-read-archive.cc | ||
| update-test-read-dwarf-output.py | ||