When coalescing ranges, sort_and_merge_range_set() unconditionally
assumes that the end of a range being folded into a preceding range
should become the end of the coalesced range. This assumption, however,
is invalid when one range is a subset of another. For example, given
ranges 1-5 and 2-3 added via range_set_append_unsafe(),
sort_and_merge_range_set() incorrectly coalesces them to range 1-3
rather than the correct union range 1-5. Fix this bug.
Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The funny range assignment in process_all_files() had me sidetracked
while investigating what led to the previous commit. Let's improve
the comments.
Signed-off-by: Thomas Rast <trast@inf.ethz.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
line_log_data has held a diff_filespec* since the very early versions
of the code. However, the only place in the code where we actually
need the full filespec is parse_range_arg(); in all other cases, we
are only interested in the path, so there is hardly a reason to store
a filespec. Even worse, it causes a lot of redundant ->spec->path
pointer dereferencing.
And *even* worse, it caused the following bug. If you merge a rename
with a modification to the old filename, like so:
* Merge
| \
| * Modify foo
| |
* | Rename foo->bar
| /
* Create foo
we internally -- in process_ranges_merge_commit() -- scan all parents.
We are mainly looking for one that doesn't have any modifications, so
that we can assign all the blame to it and simplify away the merge.
In doing so, we run the normal machinery on all parents in a loop.
For each parent, we prepare a "working set" line_log_data by making a
copy with line_log_data_copy(), which does *not* make a copy of the
spec.
Now suppose the rename is the first parent. The diff machinery tells
us that the filepair is ('foo', 'bar'). We duly update the path we
are interested in:
rg->spec->path = xstrdup(pair->one->path);
But that 'struct spec' is shared between the output line_log_data and
the original input line_log_data. So we just wrecked the state of
process_ranges_merge_commit(). When we get around to the second
parent, the ranges tell us we are interested in a file 'foo' while the
commits touch 'bar'.
So most of this patch is just s/->spec->path/->path/ and associated
management changes. This implicitly fixes the bug because we removed
the shared parts between input and output of line_log_data_copy(); it
is now safe to overwrite the path in the copy.
There's one only somewhat related change: the comment in
process_all_files() explains the reasoning behind using 'range' there.
That bit of half-correct code had me sidetracked for a while.
Signed-off-by: Thomas Rast <trast@inf.ethz.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The existing code was too defensive, and would trigger the assert in
range_set_append() if the user gave overlapping ranges.
The intent was always to define overlapping ranges as just the union
of all of them, as evidenced by the call to sort_and_merge_range_set().
(Which was already used, unlike what the comment said.)
Fix by splitting out the meat of range_set_append() to a new _unsafe()
function that lacks the paranoia. sort_and_merge_range_set will fix
up the ranges, so we don't need the checks there.
Signed-off-by: Thomas Rast <trast@inf.ethz.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
lookup_line_range() is a good place to check that the range sets
satisfy the invariants: they have been computed and set in earlier
iterations, and we now start working with them.
Signed-off-by: Thomas Rast <trast@inf.ethz.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
So far log -L only used the implicit diff filtering by pathspec. If
the user specifies -M, we cannot do that, and so we simply handed the
whole diff queue (which is approximately 'git show --raw') to
diffcore_std().
Unfortunately this is very slow. We can optimize a lot if we throw
out files that we know cannot possibly be interesting, in the same
spirit that the pathspec filtering reduces the number of files.
However, in this case, we have to be more careful. Because we want to
look out for renames, we need to keep all filepairs where something
was deleted.
This is a bit hacky and should really be replaced by equivalent
support in --follow, and just using that. However, in the meantime it
speeds up 'log -M -L' by an order of magnitude.
Signed-off-by: Thomas Rast <trast@student.ethz.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This new syntax finds a funcname matching /pattern/, and then takes from there
up to (but not including) the next funcname. So you can say
git log -L:main:main.c
and it will dig up the main() function and show its line-log, provided
there are no other funcnames matching 'main'.
Signed-off-by: Thomas Rast <trast@student.ethz.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is a rewrite of much of Bo's work, mainly in an effort to split
it into smaller, easier to understand routines.
The algorithm is built around the struct range_set, which encodes a
series of line ranges as intervals [a,b). This is used in two
contexts:
* A set of lines we are tracking (which will change as we dig through
history).
* To encode diffs, as pairs of ranges.
The main routine is range_set_map_across_diff(). It processes the
diff between a commit C and some parent P. It determines which diff
hunks are relevant to the ranges tracked in C, and computes the new
ranges for P.
The algorithm is then simply to process history in topological order
from newest to oldest, computing ranges and (partial) diffs. At
branch points, we need to merge the ranges we are watching. We will
find that many commits do not affect the chosen ranges, and mark them
TREESAME (in addition to those already filtered by pathspec limiting).
Another pass of history simplification then gets rid of such commits.
This is wired as an extra filtering pass in the log machinery. This
currently only reduces code duplication, but should allow for other
simplifications and options to be used.
Finally, we hook a diff printer into the output chain. Ideally we
would wire directly into the diff logic, to optionally use features
like word diff. However, that will require some major reworking of
the diff chain, so we completely replace the output with our own diff
for now.
As this was a GSoC project, and has quite some history by now, many
people have helped. In no particular order, thanks go to
Jakub Narebski <jnareb@gmail.com>
Jens Lehmann <Jens.Lehmann@web.de>
Jonathan Nieder <jrnieder@gmail.com>
Junio C Hamano <gitster@pobox.com>
Ramsay Jones <ramsay@ramsay1.demon.co.uk>
Will Palmer <wmpalmer@gmail.com>
Apologies to everyone I forgot.
Signed-off-by: Bo Yang <struggleyb.nku@gmail.com>
Signed-off-by: Thomas Rast <trast@student.ethz.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
