Summary:
Depends on D21051. Ref T13504. Ref T13209. This very old Mercurial code uses "X=Y hg ..." on Linux and "set X=Y & hg ..." on Windows.
The latter construct no longer works because we bypass the shell. The former construct is obsolete.
Additionaly, delete some ancient "branch merge" code which has no callers.
Test Plan: Created a diff in a Mercurial repository on Linux. I minimally vetted this on Windows since I don't have a "hg + Windows" environment at the moment.
Maniphest Tasks: T13504, T13209
Differential Revision: https://secure.phabricator.com/D21052
Summary:
Depends on D20986. Ref T13395. This //mostly// collapses the entire "experimental" branch into "master".
I plan to change the "Ref/Hardpoint" pattern to become future oriented, but this is more steps forward than sideways.
Test Plan: Ran various `arc` workflows.
Maniphest Tasks: T13395
Differential Revision: https://secure.phabricator.com/D20987
Summary: Ref T13481. Some older versions of Git appear to not support "--" in these commands. Just drop it. This can lead to ambiguous results with certain obviously-silly remote names, but doesn't appear to lead to anything dangerous.
Test Plan: Will followup with user on ancient Git.
Maniphest Tasks: T13481
Differential Revision: https://secure.phabricator.com/D20952
Summary: Fixes T13481. When identifying the URI for a remote, fall back from "git remote get-url" to "git ls-remote --get-url" to "git config remote.<name>.url" based on command output and version tests.
Test Plan: Ran `arc land --hold`, rigged the subcommands to fail to try all three fallbacks, ran `arc land --hold --remote asdfasdf` to get an explicit failure.
Maniphest Tasks: T13481
Differential Revision: https://secure.phabricator.com/D20950
Summary:
Ref T13432. Git has a "diff.suppressBlankEmpty" config option which makes it emit nonstandard diffs with trimmed trailing whitespace on unchanged blank lines.
Currently, we don't parse these diffs correctly. Even if we do in the future, emitting a more standard diff is desirable.
Explicitly disable this option when executing "git diff" so we build more standard diffs.
Test Plan:
- Configured this option.
- Modified a file with a blank line in it without changing the blank line, got this goofy display diff:
{F6985234}
- Applied patch, rediffed the same change, saw "-c diff.suppressBlankEmpty" in "--trace" output and got this sensible diff:
{F6985235}
Maniphest Tasks: T13432
Differential Revision: https://secure.phabricator.com/D20877
Summary:
See https://discourse.phabricator-community.org/t/arc-not-supporting-git-2-17-1/.
When treating it a large file binary, we try to get the "old" and "new" content using `git ls-tree` and `cat-file`.
If the file is new or deleted, there is no old file, so we try to work with filename `null`.
Under git < 2.17.1, that gets treated as `git ls-tree -- .`, which falls in the next condition under "no such path".
In git 2.18, etc, this is an error.
Explicitly bail out if there is no filename.
Test Plan: Add a new, large (>4Mb) file, arc diff.
Reviewers: epriestley, #blessed_reviewers
Reviewed By: epriestley, #blessed_reviewers
Subscribers: Korvin
Differential Revision: https://secure.phabricator.com/D19513
Summary:
Filenames are last in `git status --porcelain=2` lines; they
are not escaped in any way, despite the fields being
whitespace-delimited. `explode` thus happily chops apart filenames
with spaces in them, causing later git operations to operate only on
the filename up to the first space.
Split the lines into the right number of elements -- in all cases,
this is one more than the index we're using, since filenames come last.
Test Plan:
Altering a file with a space in its path, and running `arc diff -a`.
Added tests.
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: Korvin, epriestley
Differential Revision: https://secure.phabricator.com/D19389
Summary:
Currently, `arc` on `git` uses the following commands to examine the
state of the working tree and history; example times for a no-op diff in a
165k-file working tree are also shown:
```
1) git diff --no-ext-diff --no-textconv --submodule=short --raw 'fb062d4ecce5d9c1786b7bfc8a0dedf6b11fdd96' --
= 1,722,514 us
2a) git diff --no-ext-diff --no-textconv --submodule=short --raw 'HEAD' --
= 1,715,507 us
2b) git ls-files --others --exclude-standard
= 2,359,202 us
3) git diff-files --name-only
= 1,333,274 us
```
Steps (2a) and (2b) are run concurrently; this results in a total elapsed
wallclock time of approximately 5.4 seconds. This is inefficient -- all four of
the above steps must both load the index and examine the working copy, which may
be slow operations when large repositories are used. Additionally, none of the
effort of those stat calls on the working tree, or load time of the index, is
shared across the processes.
Step (1) is called from `getCommitRangeStatus`, which was split out in D4095; it
is currently never called on its own, only ever from `getWorkingCopyStatus`,
where it it combined with `getUncommittedStatus`. The current behavior of the
method is to return the set of changes //either// in local commits //or//
uncommitted in the working tree, which duplicates work that
`getUncommittedStatus` is intended to do. Changing the behavior of this method
(in Git, and other VCSes) to only examine _committed_ status seems both inline
with the name of the method and the original description of it in D4095 -- and
also serves to make it much faster, as it is an operation that need not inspect
the working tree at all.
Steps (2a), (2b), and (3) attempt to gather the state of the working copy, and
as such are all I/O bound but must examine nearly identical data. For git
2.11.0 and higher, we can instead rely on the machine-parseable `git status
--porcelain=2` format, which provides the information from all of these commands
at once. It also allows additional performance improvements, as `git status`
has been the focus of several optimizations in the latest versions of git (the
untracked cache and fsmonitor services, for instance), which are not available
in the lower-level `diff`, `ls-files`, and `diff-files` commands.
This has the added benefit of fixing a bug noticed in T9455, in that uncommitted
or unstaged changes in modules can now be detected, regardless of if they also
have changed their base commit. It further resolves a bug where `.gitmodules`
appeared to have unstaged changes, when in reality the unstaged changes were in
submodules elsewhere in the tree.
For backwards compatibility with versions of git < 2.11.0, the old code is left
in place. It is possible that the simpler output from v1 of `git status
--porcelain` would also suffice for some of the above benefits, but the payoff
of parsing yet another format is deemed insufficient; users wishing improved
performance should simply upgrade `git`.
Alltogether, these result in the following, for a no-op diff in a
165k-working-file tree:
```
1) git diff --no-ext-diff --no-textconv --submodule=short --raw 'fb062d4ecce5d9c1786b7bfc8a0dedf6b11fdd96' HEAD --
= 9,227 us
2) git status --porcelain=2 -z
= 739,964 us
```
...for a total of 749ms, an improvement of 4.7s.
Depends on D18841.
Test Plan: Existing tests.
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: Korvin, epriestley
Differential Revision: https://secure.phabricator.com/D18842
Test Plan:
Removed a submodule with `diff.submodule` set to `log`, saw
`arc diff` error; with this change, it no longer does.
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: epriestley
Maniphest Tasks: T10881
Differential Revision: https://secure.phabricator.com/D17327
Summary:
`git ls-remote` has an unusual way to indicate a URL was not
found: echoing back user input
```
$ git ls-remote --get-url does_not_exist
does_not_exist
$ echo $?
0
```
`getRemoteURI` handles checking for remotes other than 'origin', but
the error handling always matched against the string 'origin'
regardless of remote name.
Test Plan:
With a git config along the lines of:
```
[remote "my_special_name"]
url = ssh://secure@secure.phabricator.com/diffusion/ARC/arcanist.git
fetch = +refs/heads/*:refs/remotes/my_special_name/*
[branch "master"]
remote = github
merge = refs/heads/master
[remote "github"]
# url = git@github.com:phacility/arcanist.git
fetch = +refs/heads/*:refs/remotes/github/*
```
and running in a branch tracking `master` (github). `arc which` would
(without this diff) show:
```
The remote URI for this working copy is "github".
```
With this diff, `arc which` correctly shows:
```
Unable to determine the remote URI for this repository.
```
When diffing against a tracking branch with a propertly configured
remote (the happy path), `arc which` still correctly identifies the
remote URI:
```
The remote URI for this working copy is
"ssh://secure@secure.phabricator.com/diffusion/ARC/arcanist.git".
```
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: Korvin, chad, epriestley
Differential Revision: https://secure.phabricator.com/D17110
Summary:
`parseGitRawDiff` dealt with the `A`, `M`, and `D` status flags from
`git diff --raw`, for file additions, modifications, and deletions
respectively. However, it failed to cope with `C` and `R` flags, for
copies and renames. Git version 2.9 and above default to resolving
renames, even in `git diff --raw` output, making this lack of support
only salient now (though users with Git's `diff.rename` set
encountered it previously).
Those two flags differ from the other three in that they offer both the
source and destination filename, separated by a tab. As
`parseGitRawDiff` was not aware of this property, it returned a
"filename" of `"oldfile\tnewfile"`. This is surfaced in several
places, including as passed to linters as a filename to check.
Needless to say, this file is nearly guaranteed to never exist on
disk.
Detect both the `C` and `R` flag types, and generate either a file
addition, or a pair of addition/deletion entries.
Test Plan:
Renamed a file, with a linter that printed each file it was
called with.
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: jboning, Korvin
Differential Revision: https://secure.phabricator.com/D16387
Summary:
This is mostly just a personal quality-of-life fix. I run this command fairly often and having it return a little faster is nice.
This replaces a `git show` for each individual branch with a big `git for-each-ref` which we were already running anyway. This is quite a bit faster.
This command also occasionally hangs or segfaults for me while executing the huge pile of subprocesses. This is unreliable to reproduce, probably some bug in some PHP extension I have, and likely hard to narrow down, and this approach is better in every way anyway.
Test Plan:
- Ran `arc branch` in Git, observed faster output (in my `phabricator/`, about 2000ms -> 1200ms).
- Ran `arc feature` in Mercurial.
Reviewers: chad
Reviewed By: chad
Differential Revision: https://secure.phabricator.com/D15735
Summary:
Fixes T9953.
- "-c" was introduced in 1.7.2.
- "--no-color" has existed forever as far as I can tell.
- "--no-column" was introducd in 1.7.11, but there was nothing that needed to be disabled before that (hopefully).
Test Plan:
- Ran `arc which --trace` and observed a reasonable `git branch` command with correct output.
- Ran `arc which --trace` with a faked older Git version, observed command omit `--no-column`.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T9953
Differential Revision: https://secure.phabricator.com/D14735
Summary:
Instead of blindly assuming that "origin" is the repository that
arcanist should communicate with, use the remote that is configured
for the branch in git.
Test Plan:
Used `arc which` with a branch with no upstream, an
origin/master upstream, and an upstream/master upstream -- the last of
which is being used to create and land this diff.
Reviewers: epriestley, #blessed_reviewers
Reviewed By: epriestley, #blessed_reviewers
Subscribers: joshuaspence, Korvin
Differential Revision: https://secure.phabricator.com/D14530
Summary:
Ref T9661. I need to reuse this to fix the complex workflow described in T9661 where we need to follow multiple paths to the upstream and cascade updates across them.
Pull the logic into a separate class to make this easier and less copy/pastey.
This shouldn't change any behavior.
Test Plan: Ran `arc land --preview` from detached head, remote-tracking branch, non-tracking branch, local-tracking branch. Selection of target/remote seemed correct in all cases.
Reviewers: chad
Reviewed By: chad
Subscribers: edibiase
Maniphest Tasks: T9661
Differential Revision: https://secure.phabricator.com/D14360
Summary:
Fixes T9543. Fixes T9658. Ref T3855.
Major functional change is that you can have a sequence of branches like:
origin/master -> notmaster -> feature1
...where they track each other, but you named your local master something else. Currently, we resolve only one level of upstreams, so we try to land onto "notmaster" in this case, which is wrong.
Instead, keep resolving upstreams until we either hit a cycle, don't have another upstream to look at, or find someting in a remote. In this case we'll eventually find "origin/master" and select "origin" as the remote and "master" as the target.
Other minor changes:
- Make this selection process explicit.
- Make the help 3000x longer.
Also fix a bug where we could incorrectly try to tell Differential to update awith `--preview`.
Test Plan:
- Landed from a tag.
- Landed from a tracking branch.
- Landed from an nth-degree tracking branch.
- Tried to land from a local branch with a cycle in upstreams.
- Landed with --remote and --onto.
- Read `arc help land`.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T9658, T3855, T9543
Differential Revision: https://secure.phabricator.com/D14357
Summary:
Fixes T9455. Depends on D14136. When you have a dirty submodule:
$ nano submodule/file.c # save changes
...we currently ask you to make a commit when you run `arc diff`, which is meaningless and misleading.
Instead, prompt the user separately.
This behavior isn't perfect but I think it's about the best we can do within reason.
Test Plan:
- Ran `arc diff` in a working copy with uncommitted submodule changes only, got new prompt.
- Ran `arc diff` in a working copy with submodule base commit changes only, got old (correct) prompt.
- Ran `arc diff` in a working copy with both, got only old prompt (which is incomplete, but reasonable/meaningful).
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T9455
Differential Revision: https://secure.phabricator.com/D14137
Summary:
'git ls-remote --get-url' is more correct, but younger and less
supported. This commit tempers previous optimism about its availability,
improving support for users of older git packages.
Test Plan:
* Set `git config url.xttps.insteadOf https` rewrite rule.
* Ran `arc which` with git 1.7.5 in `$PATH`, saw rewritten configured remote.
* Ran `arc which` with git 1.7.4 in `$PATH`, saw configured remote.
Reviewers: epriestley, #blessed_reviewers
Reviewed By: epriestley, #blessed_reviewers
Subscribers: Korvin
Differential Revision: https://secure.phabricator.com/D13998
Summary:
This guards against stability issues with the output format of 'git
blame' (such as git config, localization (ref T5554) or future changes).
For example, `git config blame.blankboundary true` breaks `arc cover`
before this patch.
Test Plan:
* Set `git config blame.blankboundary true` on a test repo.
* Ran `arc cover`. It failed with an exception ("Bad blame?").
* Applied this patch.
* `arc cover` works.
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: Korvin
Maniphest Tasks: T5554
Differential Revision: https://secure.phabricator.com/D13993
Summary:
Ref T5554. Both the current branch name (if on a branch), as well as the
list of all local branches, can be retrieved without having to parse the
output from "git branch".
Unfortunately, there seems to be no git plumbing for "get list of
branches containing this commit" yet.
(see http://marc.info/?l=git&m=141408477614635&w=2)
For that case, this commit whitelists the output from "git branch" using
the known valid branch names from "git for-each-ref".
Test Plan:
Set up a test repo with this structure:
```
| * Commit B1, on branch "subfeature"
| /
| * Commit A1, on branch "feature"
|/
* Commit M1, on branch "master"
|
```
In `subfeature`, I tried:
* `arc which --base 'git:branch-unique(master)'`
* `arc feature`
After that, I detached my HEAD (don't worry, I got better) and tried again.
Nothing looked broken.
(Tested with git 1.7.2.5 and 2.5.0.)
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: Korvin
Maniphest Tasks: T5554
Differential Revision: https://secure.phabricator.com/D13989
Summary:
Ref T5554. This makes git remote URL detection locale-agnostic.
The previously suggested `git config remote.origin.url` command does
almost the same, but does not support the URL rewriting features in
git-config (`url.<base>.insteadOf`).
This one does, although it has the unintuitive behavior of just printing
the passed remote name when the remote does not exist, or even when
called outside a git repo.
Test Plan:
* Switched to non-english locale in which git has a translation.
* Ran `arc which` on the Arcanist repo. It could not determine the remote URI.
* Applied patch, `arc which` found the URI.
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: johnny-bit, Korvin
Maniphest Tasks: T5554
Differential Revision: https://secure.phabricator.com/D13983
Summary: I found a few strings that I had missed, using a mostly-broken-but-somewhat-okay custom linter ruler (https://secure.phabricator.com/differential/diff/30988/).
Test Plan: Intense eyeballing.
Reviewers: epriestley, #blessed_reviewers
Reviewed By: epriestley, #blessed_reviewers
Subscribers: aurelijus, Korvin, epriestley
Differential Revision: https://secure.phabricator.com/D12888
Summary: Ref T5781. `git show .` works like HEAD, but that isn't what `arc browse .` means.
Test Plan: Ran `arc browse .` with a repository at a published commit.
Reviewers: chad, btrahan, avive, avivey
Reviewed By: avivey
Subscribers: epriestley, avivey
Maniphest Tasks: T5781
Differential Revision: https://secure.phabricator.com/D10197
Summary:
These are documented as being identical, but `git diff a b` works if `a` is a tree (for example, `4b825d...`, the empty tree hash), but `git diff a..b` does not.
Particularly, with the `a..b` form, `arc diff --base arc:empty` does not work. With the `a b` form, it does.
Test Plan: Ran `arc diff --base arc:empty` in a repository and got a diff.
Reviewers: btrahan, talshiri
Reviewed By: talshiri
Subscribers: epriestley
Differential Revision: https://secure.phabricator.com/D9898
Summary: We manually quote this in a couple of places. That works fine on Lunix, but does not work on Windows. Instead, explicitly parameterize the command so the correct quoting rules are applied for the OS.
Test Plan: See IRC; windows user had issues fixed by this. `arc:upstream` still works locally.
Reviewers: btrahan, chad
Reviewed By: chad
Subscribers: epriestley
Differential Revision: https://secure.phabricator.com/D9868
Summary: I'm pretty sure that `@group` annotations are useless now... I believe that they were originally used by Diviner?
Test Plan: Eye-balled it.
Reviewers: epriestley, #blessed_reviewers
Reviewed By: epriestley, #blessed_reviewers
Subscribers: epriestley, Korvin, aurelijus
Differential Revision: https://secure.phabricator.com/D9855
Summary:
There's some sort of race inside `git` here, where the `git diff-files` command exits with different results some of the time when run in parallel with `git ls-files` or `git diff` (running either command was sufficient to trigger the race).
Run it separately to avoid the race.
I poked around the `git` source a little bit but quickly lost interest given that the issue seems fixed and this workaround is essentially reasonable.
Test Plan: Ran test 20x in a row without failures.
Reviewers: hach-que
Reviewed By: hach-que
Subscribers: epriestley
Differential Revision: https://secure.phabricator.com/D9616
Summary:
Introducing `--head` caused us to run `git diff base..head` explicitly.
However, we can now hit this workflow:
- We resolve `HEAD` as commit `aaaa1`.
- This is cached.
- We notice dirty working copy changes and prompt the user to amend them to HEAD.
- The user accepts the amend.
- We amend, creating commit `bbbb2`.
- We dirty the commit range and reload the working copy. This //does not// dirty the cache of HEAD.
- We run `git diff`, but it uses the old cached HEAD: `git diff base..aaaa1`.
- This works fine (`aaaa1` still exists, it's just not on any branch) but produces the wrong diff (without amended changes).
To resolve this, implement the "dirty the cache when the range reloads" hook.
Also never try to amend if the user provides `--head`.
Test Plan:
Ran `arc diff --only --trace` in a working copy with a new commit and some uncommitted changes.
- Prior to this change, saw a `git diff base..aaaa1` command and the wrong diff.
- After this change, saw a `git diff base..bbbb2` command and the correct diff.
Reviewers: chad, csilvers, talshiri
Reviewed By: talshiri
Subscribers: epriestley, spicyj
Differential Revision: https://secure.phabricator.com/D9506
Summary:
This adds support for passing range of commits for arc diff. This is useful when you want to submit code reviews for past commits without mucking around with the working copy.
This will probably require changes :)
Test Plan: Tested locally, but totally need to add tests for this
Reviewers: epriestley, #blessed_reviewers
Reviewed By: epriestley, #blessed_reviewers
Subscribers: epriestley, Korvin
Differential Revision: https://secure.phabricator.com/D9369
Summary: Ran `arc lint --apply-patches --everything` over rARC, mainly to change double quotes to single quotes where appropriate. These changes also validate that the `ArcanistXHPASTLinter::LINT_DOUBLE_QUOTE` rule is working as expected.
Test Plan: Eyeballed //most// of the diff.
Reviewers: epriestley, #blessed_reviewers
Reviewed By: epriestley, #blessed_reviewers
Subscribers: epriestley, Korvin, aurelijus
Differential Revision: https://secure.phabricator.com/D9269
Summary:
Fixes T4559.
It looks like the code currently (at least partially) handles this by checking for `(no branch)`. I suspect that the behaviour of `git` has changed (I am running version 1.9.0) because I haven't figured out what state to be in to cause `git` to output `(no branch)`.
Test Plan: Ran `arc branch` when on a detached branch.
Reviewers: #blessed_reviewers, epriestley
CC: Korvin, epriestley, aran
Maniphest Tasks: T4559
Differential Revision: https://secure.phabricator.com/D8466
Summary:
Ref T4343. Continues the process of reducing the prominence of Arcanist Projects. Primarily:
- Query Phabricator to identify the working copy based on explicit configuration, or guess based on heuristics.
- Enhance `arc which` to explain the process to the user.
- The `project_id` key is no longer required in `.arcconfig`.
Minor/cleanup changes:
- Rename `project_id` to `project.name` (consistency, clarity).
- Rename `conduit_uri` to `phabricator.uri` (consistency, clairty).
- These both need documentation updates.
- Add `repository.callsign` to explicitly bind to a repository.
- Updated `.arcconfig` for the new values.
- Fix a unit test which broke a while ago when we fixed a rare definition of "unstaged".
- Make `getRepositoryUUID()` generic so we can get rid of one `instanceof`.
Test Plan:
- Ran `arc which`.
- Ran `arc diff`.
- This doesn't really change anything, so the only real risk is version compatibility breaks. This //does// introduce such a break, but the window is very narrow: if you upgrade `arc` after this commit, and try to diff against a Phabricator which was updated after yesterday (D8068) but before D8072 lands, the lookup will work so we'll add `repositoryPHID` to the `differential.creatediff` call, but it won't exist in Phabricator yet. This window is so narrow that I'm not going to try to fix it, as I'd guess there is a significant chance that no users will be affected. I don't see a clever way to fix it that doesn't involve a lot of work, either.
Reviewers: btrahan
Reviewed By: btrahan
CC: aran
Maniphest Tasks: T4343
Differential Revision: https://secure.phabricator.com/D8073
As documented,
$ git ls-files -m
fails to exclude ignored submodules. Replace it with an equivalent:
$ git diff-files --name-only
which does not suffer from this defect.
See: <https://github.com/facebook/arcanist/pull/121>
Reviewed by: epriestley
When a submodule is ignored (ignore=all in .gitconfig),
$ git ls-files -m
fails to exclude the submodule from the listing. Other commands like
$ git diff-index --name-only HEAD
exclude it just fine.
See: <https://github.com/facebook/arcanist/pull/120>
Reviewed by: epriestley
Summary:
Lingers on from D7271; Rename `ArcanistWorkingCopyIdentity.getConfig()`.
Changed all linters (Except one) to use `getConfigFromAnySource()`, because it seems to make sense.
Test Plan: arc unit --everything; arc lint in github.com:epriestley/arclint-examples.git (Except for phpcs, flake8, cpplint and csslint which I don't have installed).
Reviewers: epriestley, #blessed_reviewers
Reviewed By: epriestley
CC: chad, Korvin, epriestley, aran
Differential Revision: https://secure.phabricator.com/D7382
Summary:
Create a new class for them, pass instance around as need.
This looks like it's mostly working, but I'd like to replace the various `new ArcanistConfigurationManager()`
calls with something more suitable.
And maybe get a better name for ArcanistConfigurationManager ("Configuration" is already taken).
Test Plan: arc unit --everything, and then some.
Reviewers: epriestley, #blessed_reviewers
Reviewed By: epriestley
CC: Korvin, epriestley, aran, chad
Differential Revision: https://secure.phabricator.com/D7271
Summary:
git ls-tree gives type 'commit' for submodules.
This code-path is only used when a binary file is present in the diff.
Test Plan: arc diff with a binary file
Reviewers: epriestley
Reviewed By: epriestley
CC: Korvin, aran
Differential Revision: https://secure.phabricator.com/D6943
Summary:
Piping data around on Windows doesn't work well when it contains zany characters like "null" and "newline". Fixes T3266.
Instead of piping data into `git apply`, write to a temporary file.
Test Plan:
Ran `arc patch`, got good results.
>>> [17] <exec> $ git apply --index --reject -- '/private/var/folders/8k/c3vkmjy5335gcxdzxkhwq82w0000gn/T/7z9iea6srikoo0sc/4266-ZEyvz9'
Reviewers: btrahan, hach-que
Reviewed By: hach-que
CC: aran
Maniphest Tasks: T3266
Differential Revision: https://secure.phabricator.com/D6070
