copyalltostore(), only one caller of copytostore(), already knows
standin file name of the target largefile. Therefore, passing it to
copytostore() is more efficient than calculating it in copytostore()
or readstandin().
This will be used to centralize and encapsulate the logic to read hash
from given (filectx of) standin file. readstandin() isn't suitable for
this purpose, because there are some code paths, which want to read
hex hash directly from filectx.
Before this patch, updatestandin() takes "standin" argument, and
applies splitstandin() on it to pick out a path to largefile (aka
"lfile" or so) from standin.
But in fact, all callers already knows "lfile". In addition to it,
many callers knows both "standin" and "lfile".
Therefore, making updatestandin() take only one of "standin" or
"lfile" is inefficient.
repo['.'] is called not as "working context" but as "parent context".
In this code path, hash value of current content of file should be
compared against hash value recorded in "parent context".
Therefore, "wctx" may cause misunderstanding in this case.
This patch also renames argument of hexsha1(), not only for
readability ("data" isn't good name for file-like object), but also
for reviewability (including hexsha1() code helps reviewers to confirm
how these functions are similar).
BTW, copyandhash() has also similar logic, but it can't reuse
hexsha1(), because it writes read-in data into specified fileobj
simultaneously.
There are some code paths, which apply standin() on same value
multilpe times instead of using already standin()-ed value.
"fstandin" is common name for "path to standin file" in lfutil.py, to
avoid shadowing "standin()".
readstandin() takes "node" argument to get changectx by "repo[node]".
There are some readstandin() invocations, which use ctx.node(),
ctx.rev(), or '.' as "node" argument above, even though corresponded
changectx object is already looked up on caller side.
This patch calls readstandin() with already known changectx itself, to
avoid meaningless re-construction of changectx (indirect case via
copytostore() is also included).
BTW, copytostore() uses "rev" argument only for readstandin()
invocation. Therefore, this patch also renames it to "revorctx" to
indicate that it can take not only revision ID or so but also
changectx, for readability.
There are many isstandin() invocations before splitstandin().
The former examines whether specified path starts with ".hglf/". The
latter returns after ".hglf/" of specified path if it starts with that
prefix, or returns None otherwise.
Therefore, value returned by splitstandin() can be used for
replacement of preceding isstandin(), and this replacement can omit
redundant string comparison after isstandin().
Now that the 'vfs' classes moved in their own module, lets use the new module
directly. We update code iteratively to help with possible bisect needs in the
future.
pycompat.getenv returns os.getenvb on py3 which is not available on Windows.
This patch replaces them with encoding.environ.get and checks to ensure no
new instances of os.getenv or os.setenv are introduced.
os.getenv deals with unicodes on Python 3, so we have pycompat.osgetenv to
deal with bytes. This patch replaces occurrences on os.getenv with
pycompat.osgetenv
util.filechunkiter has been using a chunk size of 64k for more than 10 years,
also in years where Moore's law still was a law. It is probably ok to bump it
now and perhaps get a slight win in some cases.
Also, largefiles have been using 128k for a long time. Specifying that size
multiple times (or forgetting to do it) seems a bit stupid. Decreasing it to
64k also seems unfortunate.
Thus, we will set the default chunksize to 128k and use the default everywhere.
Before, we would sometimes use the default iterator over large files. That
iterator is line based and would add extra buffering and use odd chunk sizes
which could give some overhead.
copyandhash can't just apply a filechunkiter as it sometimes is passed a
genuine generator when downloading remotely.
Problem in both cases is cache in largefiles has assigned
meaning - user cache which is additional place to get/put
files. Those two function works on store - the main place
to store largefiles in the repository - .hg/largefiles and
using "cache" to describe it is misleading.
All versions of Python we support or hope to support make the hash
functions available in the same way under the same name, so we may as
well drop the util forwards.
Prior to revision 149be6a0072e, largefiles were saved in the local repository,
even if it was using the share extension. After that change, all largefiles are
now stored in the shared repository. However, the backward compatibility for
existing largefiles already placed in the local repository was never tested,
and has been broken since.
9f1a3c7b4a28 introduced support for not having a "global" user cache.
In the rare cases where the environment didn't provide the location of the
current home directory, the usercachepath function could return None.
That functionality has since bitrotten and several code paths did not correctly
check for usercachepath returning None:
$ HOME= XDG_CACHE_HOME= hg up --config extensions.largefiles=
getting changed largefiles
abort: unknown largefiles usercache location
Dropping the partial support for it is thus not really a backward compatibility
breaking change.
Thus: consistently fail early if the usercache location is unknown.
It is relevant to be able to control where the largefiles are stored and how
they propagate, but that should probably be done differently. The dysfunctional
code just gets in the way.
Replaces invocations os.path functions to methods in vfs. Unfortunately
(in my view) this makes code less readable, because instead of using
clear variable names with path it needs to replace them with vfs(..).
I need guidance how to make such transition look more readable.
For example in this patch there is example with few places with
wvfs.join(standindir), standindir before this patch was absolute
path, in this it is changed to relative because it is used also
in expression wvfs.join(standindir, pat).
Using plural form is consistent with other progress units, and "1 out of 5
revisions" sounds more correct. Also, tests don't show this, but if you have
'speed' item in progress.format config, it shows e.g. '100 revisions/sec',
which also seems better.
Previously, if the largefile was deleted at the time of a commit, the standin
was silently not updated and its current state (possibly garbage) was recorded.
The test makes it look like this is somewhat of an edge case, but the same thing
happens when an `hg revert` followed by `rm` changes the standin.
Aside from the second invocation of this in lfutil.updatestandinsbymatch()
(which is what triggers this test case), the three other uses are guarded by
dirstate checks for added or modified, or an existence check in the filesystem.
So aborting in lfutil.updatestandins() should be safe, and will avoid silent
skips in the future if this is used elsewhere.
The change in 6fce9a02f069 to handle a normal -> largefile switch was too
aggressive in preserving the original matcher names. If a largefile is
explicitly provided by the user, but only the standin exists in dirstate, then
only the standin can be committed.
There's still maybe an issue when the largefile is deleted outside of Mercurial:
$ rm large
$ hg ci -m "oops" large
large: The system cannot find the file specified
nothing changed
[1]
92117e4f6f8d improved merging of standin files referencing missing largefiles.
It did however not test or fix commits of such merges; it would abort.
To fix that, change copytostore to skip and warn about missing largefiles
with a message similar the one for failing get from remote filestores. (It
would perhaps in both cases be better to emit a more helpful warning like
"warning: standin file for large1 references 58e24f733a which can't be found in
the local store".)
To test this, make sure commit doesn't find the "missing" largefile in the global
usercache. For further testing, verify that update and status works as expected
after this.
This will also effectively backout 159c82dd6523.
If the store somehow got corrupted, users could end up in weird situations that
were very hard to recover from or lead to propagation of the corruption.
Instead, spend the extra time checking the hash when copying to the working
directory. If it doesn't match, emit a warning, and don't put wrong content in
the working directory.
Commit of corresponding normal/largefiles pairs would only commit the standin.
That is usually fine, except if either the normal file or the standin is a
remove while the other is an add. In that case it would either give duplicate
colliding entries or lose the file.
Instead, commit both filenames if one of them is a remove.
Before, when merging revisions with missing largefiles, the missing largefiles
would be fetched as a part of the merge. If that failed (for example because
the main repository temporarily was unavailable), the largefile would be left
missing. However, the next commit would abort and (seemed to) fail when
markcommitted tried to mark the standin file as normal and thus had to hash the
largefile that didn't exist. (Actually, the commit would succeed but the
largefile update that follows right after the commit transaction would abort -
quite confusing.)
To fix that, make sure that synclfdirstate only marks files as normal if they
actually exist.
The home of 'Abort' is 'error' not 'util' however, a lot of code seems to be
confused about that and gives all the credit to 'util' instead of the
hardworking 'error'. In a spirit of equity, we break the cycle of injustice and
give back to 'error' the respect it deserves. And screw that 'util' poser.
For great justice.
Previously, simply having the largefiles extension loaded without any largefiles
added would crash when amending with -I. The problem was with no files in the
matcher, the pattern list of files joined with 'standindir' was empty, and
scmutil.match() would match everything. In lfutil.composestandinmatcher(), the
match function is used to test if the file is a standin, and after getting a
false positive, proceeds to call lfutil.splitstandin(). This returns None
because it isn't a standin, which blows up when passed to rmatcher.matchfn().
Manually overriding _always in getstandinmatcher() probably isn't necessary
anymore, but we leave well enough alone on stable. This regressed in
78632d61a993.
The monkey patching in cat() can't be fixed, because it still delegates to the
original bad(). Overriding commands.cat() should go away in favor overriding
cmdutil.cat() anyway, and that matcher can be wrapped with matchmod.badmatch().
The choice between the "always" case and the other case is done in
getstandinmatcher() and the next patch will change how it's determined
based on the matcher, so let's prepare by passing in the matcher, not
just the matcher's files.
The benefit of retargeting the local store to the share source is that all
shares will always have access to the largefiles any one of them commit, even if
the user cache is deleted (which is documented to be OK to do). Further, any
push into the source (and now any shares), will likewise make the largefile(s)
visible to all related repositories.
In order to maintain compatibility with existing repos, where the largefiles
would be cached only in the local share, fallback to searching the local share
if it isn't found at the share source.
The unshare command should probably be taught to copy the source store into the
store for the repo being unshared to complete the loop.
This patch changes the test like this:
@@ -159,6 +159,5 @@
$ hg share -q src share_dst --config extensions.share=
$ hg -R share_dst update -r0
getting changed largefiles
- large: largefile $HASH not available from file:///$TESTTMP\share_dst
- 0 largefiles updated, 0 removed
+ 1 largefiles updated, 0 removed
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
The issue writeup mentions pushing a largefile from a remote repo to the main
local repo, and the largefile is then not available in any shares. Since the
push doesn't cache the largefile in $USERCACHE, the trashed $USERCACHE in this
test is equivalent.
The handful of direct uses of lfutil.storepath() merely need a single path to
read from or write to the largefile, whether or not it exists. Most callers
that care about the file existing call lfutil.findfile(), in order to fallback
from the store to the user cache.
localstore._verify() doesn't call lfutil.findfile(). This prevents redirecting
the store to the share source because the largefiles for existing repos may not
be in the source's store, so verification may fail. It can't be changed to call
findfile(), because findfile() links the file from the usercache to the local
store[1], and because it returns None instead of a path if the file doesn't
exist.
For now, this method is just a cover for lfutil.storepath(), but it will be
filled out in an upcoming patch.
[1] Maybe we shouldn't care? But on a filesystem that doesn't support
hardlinks, then verify will take a lot longer, and start to consume disk
space.