Before this patch, "rebase.concludenode()" uses "dirstate.invalidate()"
as a kind of "restore .hg/dirstate to the original status" during a failure.
But it just discards changes in memory, and doesn't actually restore
".hg/dirstate". Then, it can't work as expected, if "dirstate.write()"
is executed while processing.
This patch uses "dirstateguard" instead of "dirstate.invalidate()" to
restore ".hg/dirstate" during a failure even if "dirstate.write()" is
executed before a failure.
This patch also removes "beginparentchage()" and "endparentchange()",
because "dirstateguard" makes them useless.
This is a part of preparations to fix the issue that the recent (in
memory) dirstate isn't visible to external processes (e.g. "precommit"
hook).
After this patch, the changed dirstate becomes visible to external
"precommit" hooks during "hg rebase" in "test-largefiles-misc.t",
because "dirstateguard()" writes it out. But this content isn't yet
correct, because:
- "normal3" should be marked as "A"(dded) at committing
It is newly added in the changeset being rebased.
- but it is marked as "M"(odified)
The result of "repo.setparents()" after "dirstateguard()" isn't
yet written out before "precommit". So, merging is still in
progress for "hg status" in it.
This causes marking the file newly added on "other" branch as "A".
This will be fixed by subsequent patch.
Before this patch, releasing the store lock implies the actions below, when
the transaction is aborted:
1. "commithook()" scheduled in "localrepository.commit()" is invoked
2. "changectx.__init__()" is invoked via "self.__contains__()"
3. specified ID is examined against "repo.dirstate.p1()"
4. validation function is invoked in "dirstate.p1()"
In subsequent patches, "dirstate.invalidate()" invocations for
discarding changes are replaced with "dirstateguard", but discarding
changes by "dirstateguard" is executed after releasing the store lock:
resources are acquired in "wlock => dirstateguard => store lock" order,
and are released in reverse order.
This may cause that "dirstate.p1()" still refers to the changeset to be
rolled-back at (4) above: pushing multiple patches by "hg qpush" is
a typical case.
When releasing the store lock, such changesets are:
- not contained in "repo.changelog", if it is reloaded from
".hg/00changelog.i", as that file was already truncated by
"transaction.abort()"
- still contained in it, otherwise
(this "dirty read" problem is discussed in "Transaction Plan"
http://mercurial.selenic.com/wiki/TransactionPlan)
Validation function shows "unknown working parent" warning in the
former case, but reloading "repo.changelog" depends on the timestamp
of ".hg/00changelog.i". This causes occasional test failures.
In the case of scheduled "commithook()", it just wants to examine
whether "node ID" of committed changeset is still valid or not. Other
examinations implied in "changectx.__init__()" are meaningless.
To avoid showing the "unknown working parent" warning irregularly, this
patch uses "changelog.hasnode()" instead of "node in self" to examine
existence of committed changeset.
Since e902e55c3d0b, column headers are wrapped by <thead> element, so the first
and only <tbody> contains changelog data. I got the following error without
this patch:
Uncaught TypeError: Cannot read property 'lastElementChild' of null
scrollHandler @ mercurial.js:375
Because double backslashes are processed as a string escape sequence, '\\{'
should start the template syntax. On the other hand, r'' disables any sort
of \-escapes, so r'\{' can go either way, never start the template syntax
or always start it. I simply chose the latter, which means r'\{' is the same
as '\\{'.
This patch brings back pre-2.8.1 behavior.
The result of parsestring() is stored in templater's cache, t.cache, and then
it is parsed as a template string by compiletemplate(). So t.cache should keep
an unparsed string no matter if it is sourced from config value. Otherwise
backslashes would be processed twice.
The test vector is borrowed from 83ff877959a6.
test-histedit-edit was broken because it relied on the HGEDITOR script being
executable. Instead, lets just execute 'sh' and pass it the script to run. This
seems to be the pattern followed in other tests.
The previous behavior when archiving a subrepo 's' on Windows was to internally
name the file under it 's\file', due to the use of vfs.reljoin(). When printing
the file list from the archive on Windows or Linux, the file was named
's\\file'. The archive extracted OK on Windows, but if the archive was brought
to a Linux system, it created a file named 's\file' instead of a directory 's'
containing 'file'.
*.zip format achives seemed not to have the problem, but this was definitely an
issue with *.tgz archives.
Largefiles actually got this right, but a test is added to keep this from
regressing. The subrepo-deep-nested-change.t test was repurposed to archive to
a file, since there are several subsequent tests that archive to a directory.
The output change is losing the filesystem prefix '../archive_lf' and not
listing the directories 'sub1' and 'sub1/sub2'.
--edit-plan was completely broken from the command line because it used an old
api that was not updated (it would crash with a stack trace). Let's update it
and add tests to catch this.
The patch solves two issues:
1. id(unknown_full_hash) aborts, but id(unknown_short_hash) doesn't
2. id(40byte_tag_or_bookmark) returns tagged/bookmarked revision,
but id(non-40byte_tag_or_bookmark) doesn't
After the patch:
1. id(unknown_full_hash) doesn't abort
2. id(40byte_tag_or_bookmark) returns empty set
One case where that would happen is while trying to resolve a subrepo, if the
path to the subrepo was actually a broken symlink. This bug was exposed by an
hg-git test.
With "dash" (as "/bin/sh" on Debian GNU/Linux), command execution in
"ENV=val foo bar" style doesn't work as expect in test script files,
if "foo" is user-defined function: it works fine, if "foo" is existing
commands like "hg".
5acecb004820 introduced tests for HGPLAIN and HGPLAINEXCEPT into
test-revset.t, and all of them are in such style.
This patch doesn't:
- add explicit unsetting for HGPLAIN and HGPLAINEXCEPT
they are already introduced by 5acecb004820
- write assignment and exporting in one line
"ENV=val; export ENV" for two or more environment variables in one
line causes failure of test-check-code-hg.t: it is recognized as
"don't export and assign at once" unfortunately.
The editor launches without expanding the path with commits because the shell
does that for us.
If the path isn't an executable, the expanded path is displayed, which is
probably more useful than the unexpanded path. For example, in cmd.exe, '~'
expands to C:\Users\$user. But it expands to C:/mingw/msys/1.0/home/$user in
MinGW.
The '~' in the bug report is being expanded to a path with Windows style slashes
before being passed to shellquote() via util.shellquote(). But shlex.split()
strips '\' out of the string, leaving an invalid path in dispatch.aliasargs().
This regressed in 72640182118e.
For now, the tests need to be conditionalized for Windows (because those paths
are quoted). In the future, a more complex regex could probably skip the quotes
if all component separators are double '\'. I opted to glob away the quotes in
test-rename-merge2.t and test-up-local-change.t (which only exist on Windows),
because they are in very large blocks of output and there are way too many diffs
to conditionalize with #if directives. Maybe the entire path should be globbed
away like the following paths in each changed line. Or, letting #if directives
sit in the middle of the output as was mentioned a few months back would work
too.
Unfortunately, I couldn't figure out how to test the specific bug. All of the
'hg serve' tests have a #require serve declaration, causing them to be skipped
on Windows. Adding an alias for 'expandtest = outgoing ~/bogusrepo' prints the
repo as '$TESTTMP/bogusrepo', so the test runner must be changing the
environment somehow.
ui.plain() is supposed to disable config options that change the UI to the
detriment of scripts. As the test demonstrates, revset aliases can actually
override builtin ones, just like command aliases. Therefore I believe this is a
bugfix and appropriate for stable.
This avoids the following error that happened if base revision of bundle file
was hidden. bundlerevlog needs it to construct revision texts from bundle
content as revlog.revision() does.
File "mercurial/context.py", line 485, in _changeset
return self._repo.changelog.read(self.rev())
File "mercurial/changelog.py", line 319, in read
text = self.revision(node)
File "mercurial/bundlerepo.py", line 124, in revision
text = self.baserevision(iterrev)
File "mercurial/bundlerepo.py", line 160, in baserevision
return changelog.changelog.revision(self, nodeorrev)
File "mercurial/revlog.py", line 1041, in revision
node = self.node(rev)
File "mercurial/changelog.py", line 211, in node
raise error.FilteredIndexError(rev)
mercurial.error.FilteredIndexError: 1
There were 2 test failures in 3.4-rc when running test-hook.t with the
largefiles extension enabled. For context, the first is a commit hook:
@@ -618,9 +621,9 @@
$ echo 'update = hg id' >> .hg/hgrc
$ echo bb > a
$ hg ci -ma
- 223eafe2750c tip
+ d3354c4310ed+
$ hg up 0
- cb9a9f314b8b
+ 223eafe2750c+ tip
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
make sure --verbose (and --quiet/--debug etc.) are propagated to the local ui
In both cases, largefiles acquires the wlock before calling into core, which
also acquires the wlock. The first case was fixed in 4100e338a886 by ensuring
the hook only runs after the lock has been fully released. The full release is
important, because that is what writes dirstate to the disk, allowing external
hooks to see the result of the update. This simply changes how the update hook
is called, so that it too is deferred until the lock is finally released.
There are many uses of mergemod.update(), but in terms of commands, it looks
like the following commands take wlock while calling mergemod.update(), and
therefore will now have their hook fired at a later time:
backout, fetch, histedit, qpush, rebase, shelve, transplant
Unlike the others, fetch immediately unlocks after calling update(), so for all
intents and purposes, its hook invocation is not deferred (but the external hook
still sees the proper state).
The current bundle2 processing was capturing all output. This is nice as it
provide better meta data about what output what, but this was changing two
things:
1) adding a prefix "remote: " to "other" output during local push (issue4613)
2) local and ssh push does not provide real time output anymore (issue4615)
As we are unsure about what form should be used in (1) and how to solve (2) we
disable output capture in this two cases. Output capture can be forced using an
experimental option.
This commit modifies, adds, renames, removes files all at once. It
will be mostly interesting for an upcoming test relating to the status
log template.
We are about to record more complex time-related data, which will require
repeated changes of every loop touching times data. That will also extend such
lines to a point where things become too long. Instead, we iterate on each entry
and expand values in the loops. We keep intermediate variables in most cases to
preserve readability.
The loop producing json data does a strange inversion for no obvious reason. We
preserved that for now and will fix it in another changeset.
e926f2ef639a fixed the issue of double escapes, but it made the following
template fail with syntax error because of <\">. Strictly speaking, <\">
appears to be invalid in non-string part, but we are likely to escape <">
if surrounded by quotes, and we are used to write such templates by trial
and error.
[templates]
sl = "{tags % \"{ifeq(tag,'tip','',label('log.tag', ' {tag}'))}\"}"
So, for backward compatibility between 2.8.1 and 3.4, a single backslash
before quotation mark is stripped only in quoted template. We don't care
for <\"> in string literal in quoted template, which never worked as expected
before.
template result
--------- ------------------------
{\"\"} parse error
"{""}" {""} -> <>
"{\"\"}" {""} -> <>
{"\""} {"\""} -> <">
'{"\""}' {"\""} -> <">
"{"\""}" parse error (don't care)
This methods is needed internally by utilities like __repr__. I do not see any
harm in having it declared. Any actual attempt to use it to call a test would
crash, but at least random debug print in the test runner will no longer crash.
Another approach would have been to redefine the __repr__ function, but I think
it a good thing to have some respect for the base class API.
d6f3cca7f3a5 is adding a new tar call but forgot to filter out the
pax_global_header file that some tar versions write. This is similar to what
happened in 686775e575f3.
Unix utilities like tar will happily prefix the files it packs with './', but
annoyingly, Windows Explorer will not show these packed files when it opens the
archive. Since there doesn't seem to be a point in including './' in the path
names, just drop it. The default 'hg archive' prefix is the basename of the
archive, so specifying '.' allows for that default to be disabled completely.
'git diff-index' by default does not recurse into subdirectories
when changes are found. As a result, the directory is shown as changed,
rather than the files in this directory.
Adding '-r' results in recursing until the blobs themselves are checked.
With many commands accepting a '-S' or an explicit path to trigger recursing
into subrepos, it seems that --hidden needs to be propagated too.
Unfortunately, many of the subrepo layer methods discard the options map, so
passing the option along explicitly isn't currently an option. It also isn't
clear if other filtered views need to be propagated, so changing all of those
commands may be insufficient anyway.
The specific jam I got into was amending an ancestor of qbase in a subrepo, and
then evolving. The patch ended up being hidden, and outgoing said it would only
push one unrelated commit. But push aborted with an 'unknown revision' that I
traced back to the patch. (Odd it didn't say 'filtered revision'.) A push with
--hidden worked from the subrepo, but that wasn't possible from the parent repo
before this.
Since the underlying problem doesn't actually require a subrepo, there's
probably more to investigate here in the discovery area. Yes, evolve + mq is
not exactly sane, but I don't know what is seeing the hidden revision.
In lieu of creating a test for the above situation (evolving mq should probably
be blocked), the test here is a marginally useful case where --hidden is needed
in a subrepo: cat'ing a file in a hidden revision. Without this change, cat
aborts with:
$ hg --hidden cat subrepo/a
skipping missing subrepository: subrepo
[1]
When passing a --rev, 'hg incoming -S' previously suffered from the same output
truncation or abort that was fixed for 'hg outgoing -S' in the previous patch,
for the same reasons.
Unlike push, subrepos are currently only pulled when the outer repo is updated,
not when the outer repo is pulled. That makes matching 'hg pull' behavior
impossible. Listing all incoming csets in the subrepo seems like the most
useful behavior, and is consistent with 'hg outgoing -S'.
The previous behavior didn't reflect what would actually be pushed- push will
ignore --rev and --branch in the subrepo and push everything. Therefore,
'push -r {rev}' would not list everything, unless {rev} was also the revision of
the subrepo's tip. Worse, if a hash was passed in, the command would abort
because that hash would either not be in the outer repo or not in the subrepo.
While fixing issue4304: "record: allow editing new files" we introduced
changes in record/crecord. These changes need to be matched with changes in any
command using record. Shelve is one of these commands and the changes have
not been made for this release. Therefore, shelve -i should be an experimental
feature for this release.
In my latest change on record (edit newly added file), I forgot the
repo.wjoin() so that record was not computing the paths properly to delete
the backups and was crashing.
The '' that is used to represent the state of a not-yet-committed
subrepo cannot be written to the file, because the code that parses
the file splits on ' ' and expects two parts.
Given that the .hgsubstate file is automatically rewritten on commit, it seems
a little strange that the file is written out during a merge.
This regressed in 39be809de631, in what looks like an unrelated change.
It seems sufficient to pass 'ignoremissing=True', but the restored try/except
has been there for six years since 3f24f7324cd8, so this seems safer for now.
Note that renaming directories in the filename doesn't appear to work- not sure
if this would end up throwing a different type of error when that is fixed.