Because of the way filenodes are computed, you can have multiple changesets
"introducing" the same file revision. For example, in the changeset graph
below, changeset 2 and 3 both change a file -to- and -from- the same content.
o 3: content = new
|
| o 2: content = new
|/
o 1: content = old
In such cases, the file revision is create once, when 2 is added, and just reused
for 3. So the file change in '3' (from "old" to "new)" has no linkrev pointing
to it). We'll call this situation "linkrev-shadowing". As the linkrev is used for
optimization purposes when walking a file history, the linkrev-shadowing
results in an unexpected jump to another branch during such a walk.. This leads to
multiple bugs with log, annotate and rename detection.
One element to fix such bugs is to ensure that walking the file history sticks on
the same topology as the changeset's history. For this purpose, we extend the
logic in 'basefilectx.parents' so that it always defines the proper changeset
to associate the parent file revision with. This "proper" changeset has to be an
ancestor of the changeset associated with the child file revision.
This logic is performed in the '_adjustlinkrev' function. This function is
given the starting changeset and all the information regarding the parent file
revision. If the linkrev for the file revision is an ancestor of the starting
changeset, the linkrev is valid and will be used. If it is not, we detected a
topological jump caused by linkrev shadowing, we are going to walk the
ancestors of the starting changeset until we find one setting the file to the
revision we are trying to create.
The performance impact appears acceptable:
- We are walking the changelog once for each filelog traversal (as there should
be no overlap between searches),
- changelog traversal itself is fairly cheap, compared to what is likely going
to be perform on the result on the filelog traversal,
- We only touch the manifest for ancestors touching the file, And such
changesets are likely to be the one introducing the file. (except in
pathological cases involving merge),
- We use manifest diff instead of full manifest unpacking to check manifest
content, so it does not involve applying multiple diffs in most case.
- linkrev shadowing is not the common case.
Tests for fixed issues in log, annotate and rename detection have been
added.
But this changeset does not solve all problems. It fixes -ancestry-
computation, but if the linkrev-shadowed changesets is the starting one, we'll
still get things wrong. We'll have to fix the bootstrapping of such operations
in a later changeset. Also, the usage of `hg log FILE` without --follow still
has issues with linkrev pointing to hidden changesets, because it relies on the
`filelog` revset which implement its own traversal logic that is still to be
fixed.
Thanks goes to:
- Matt Mackall: for nudging me in the right direction
- Julien Cristau and Rémi Cardona: for keep telling me linkrev bug were an
evolution show stopper for 3 years.
- Durham Goode: for finding a new linkrev issue every few weeks
- Mads Kiilerich: for that last rename bug who raise this topic over my
anoyance limit.
This removes an optimization that was introduced in 5a644704d5eb but was too
aggressive - as indicated by how it changed test-mq-merge.t .
We are walking filelogs to find copy sources and we can thus not be sure to hit
the base revision and find the renamed file there - it could also be in the
first ancestor of the base ... in the filelog.
We are walking the filelog and can thus not easily know when we hit the first
ancestor of the base revision and which filename to look for there. Instead, we
use _findlimit like mergecopies do: The lower bound for how far we have to go
is found from the lowest changelog revision that is an ancestor of only one of
the compared revisions. Any filelog ancestor with a revision number lower than
that revision will be the ancestor of both compared revisions, and there is
thus no reason to go further back than that.
The inverse of a rename is a rename, but the inverse of a copy is not a copy.
Presenting it as such -- in particular, stuffing it into the same dict as real
copies -- causes bugs because other code starts believing the inverse copies
are real.
The only test whose output changes is test-mv-cp-st-diff.t. When a backwards
status -C command is run where a copy is involved, the inverse copy (which was
hitherto presented as a real copy) is no longer displayed.
Keeping track of inverse copies is useful in some situations -- composability
of diffs, for example, since adding "a" followed by an inverse copy "b" to "a"
is equivalent to a rename "b" to "a". However, representing them would require
a more complex data structure than the same dict in which real copies are also
stored.
The existing copy detection API was designed with merge in mind and
was ill-suited for doing status/diff. The new pathcopies
implementation gives more accurate, easier to use results for
comparing two revisions, and is much simpler to understand.
Test notes:
- test-mv-cp-st.t results finds more renames in the reverse direction now
- test-mq-merge.t was always wrong and duplicated a copy in diff that
was already present in one of the parent revisions
Globbing is usually used for filenames, so on windows it is reasonable and very
convenient that glob patterns accepts '\' or '/' when the pattern specifies
'/'.
