* include/abg-diff-utils.h (point::set): New overload..
(point::{add, operator<, operator>, operator<=, operator>=}): New
methods.
(point::operator!=): Constify.
(point::operator==): Constify. Cleanup.
(point::operator=): Keep emptiness.
(class snake): New class definition
(d_path_vec::{over_bounds, offset}): New methods.
(d_path_vec::check_index_against_bound): Don't take a bound
parameter anymore. Use the new over_bound method above. Fix up
error reporting.
(d_path_vec::d_path_vec): Fix d_path_vec size allocation.
(d_path_vec::operator[]): Use the d_path_vec::at method to check
all accesses against the bounds. This is slower, but at least we
can expect to have something that is more robust. We can remove
the bound checking later when we are sure the code has been tested
enough. Also use the new offset() method.
(d_path_vec::at): Take long long.
(ends_of_furthest_d_paths_overlap): Constify input parameters.
(end_of_fr_d_path_in_k, end_of_frr_d_path_in_k_plus_delta): Take
an instance of the new snake in parameter, rather than a bare end
point that wasn't carrying enough information about the snake.
Record the snake which consists of up to four points: a begin
point, an intermediate point, a diagonal start point and an end
point. Return that snake upon successful completion.
(compute_middle_snake): Take an instance of snake, rather than the
two points that were supposed to represent a snake and with which
we were loosing information before. Revisit/simplify the logic of
this function; this literally goes forward or in reverse, gets the
resulting snake returned by the end_of_fr_d_path_in_k and
end_of_frr_d_path_in_k_plus_delta functions, detect if these snakes
overlap and just return the current snake. Much simpler. The
caller now gets a snake, which has much more information than the
previous snake approximation made of just two points. Bonus
point, this follows almost to the word, what the paper says.
(maybe_record_match_point, find_snake_start_point): Remove these
as there are not used by compute_middle_snake anymore.
(print_snake, ses_len): Update these to take/handle a snake.
(snake_end_points): New declaration.
(compute_diff): When we are getting an empty first sequence, this
means that we are inserting the second sequence *before* the
beginning of the first sequence; keep this information by setting
the insertion point index to -1, rather than zero. Update this to
get/handle snakes, rather than free points vaguely representing
snakes. Now that compute_middle_snake returns real snakes, handle
the information we are getting. Basically for edit scripts of
length equal to 1, as the snake carries all the necessary
information about the non-diagonal edge (as well as the diagonal
edges), we (can) now precisely update the current edit script (as
well as the longest common sub-sequence). For edit scripts of
length greater than 1, better at which points to divide the
problem and consequently, at which points to conquer it back --
better following The Paper to the letter.
(display_edit_script): Update this for the use of instances of
snake.
* src/abg-diff-utils.cc (ends_of_furthest_d_paths_overlap): Update
for constification of inputs.
(snake_end_points): Define new function.
(compute_middle_snake): Adapt for the taking an instance of snake.
* tests/test-diff2.cc (main): Update for using instances of snake.
* tests/test-core-diff.cc: Add new tests.
* tests/data/test-core-diff/report0.txt: Update for output
adaptation.
* tests/data/test-core-diff/report6.txt: Likewise.
* tests/data/test-core-diff/report7.txt: Likewise.
* tests/data/test-core-diff/report8.txt: New test data.
* tests/data/test-core-diff/report9.txt: Likewise.
* tests/data/test-core-diff/report10.txt: Likewise.
* tests/data/test-core-diff/report11.txt: Likewise.
* tests/data/test-core-diff/report12.txt: Likewise.
* tests/data/test-core-diff/report3.txt: Likewise.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
* include/abg-diff-utils.h (compute_middle_snake): After the
overlap determination happened, finding the middle snake can
require keep on building the current path until the "end". The
end meaning reaching the max of D. And that max is (M + N)/2 + 1.
In the extreme cases were middle snake was on the very last step
(M + N) + 1, we were not finding the middle snake. Fix this.
(compute_diff): When d == 1 and the first edge on the edit graph
is a non-diagonal edge and when a_base != a_begin, we were failing
to properly initialize x,y to find that non-diagonal edge. Also
we were failing to correctly compute the size of the sequence.
Fix these.
* tests/test-core-diff.cc: Add a new regression test for the two
cases above.
* tests/data/test-core-diff/report7.txt: New reference data for
the new regression test.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
* include/abg-diff-utils.h (point::{operator!=,operator==}): New
operators.
(end_of_fr_d_path_in_k, end_of_frr_d_path_in_k_plus_delta): Allow
the initial point (-1,-1) that is not a point addressing elements
of the input sequences, but that is the starting point of the
forward paths and the ending point of reverse paths in the "Linear
Refinement" of the algorithm.
(is_match_point, maybe_record_match_point)
(find_snake_start_point): New functions.
(find_last_snake_in_path): Remove this. It's not used anymore.
(compute_middle_snake): Allow checking for overlapping paths even
on points that are outside of the edit graph boundaries. Once the
overlap is detected, if a non-empty snake has been seen already,
report it as the middle snake. Otherwise, keep building the path
until the end and report the last snake encountered as the middle
snake. Add comments.
(compute_diff): For the d == 1 case, fix the logic of the finding
the non-diagonal edge. Fix typos. Add comments.
(display_edit_script): Fix report glitches.
* tests/data/test-core-diff/report3.txt: Update as per the report
glitch above.
* tests/data/test-core-diff/report4.txt: Likewise.
* tests/data/test-core-diff/report5.txt: Likewise.
* tests/data/test-core-diff/report6.txt: New reference report for
a new test.
* tests/test-core-diff.cc: Add a new test for negative delta.
Signed-off-by: Dodji Seketeli <dodji@redhat.com>
