Summary:
Ref T11968. Inches toward the new ref/hardpoint code by introducing the modern refs as "RefPro" objects and supporting an "arc inspect <object>" to load objects and hardpoints.
This doesn't impact any existing runtime behavior.
Test Plan: Ran "arc inspect [--all] commit(...)", got hardpoint queries and yield-based data fetching.
Maniphest Tasks: T11968
Differential Revision: https://secure.phabricator.com/D21078
Summary:
Depends on D21074. Ref T13490. Ref T11968. Before toolsets, Arcanist has a "WorkingCopyIdentity" object. After toolsets, it has a "WorkingCopy" object.
Most workflows don't access either one, so make a slightly-breaking API change to simplify the transition.
- "getWorkingCopy()" now returns a modern object (a "WorkingCopy").
- "getWorkingCopyIdentity()" now returns an older object (a "WorkingCopyIdentity").
This isn't backward-compatible, but out-of-date code should get an explicit failure with clear resolution steps.
Test Plan: Ran "arc lint", "arc branch", and a few other commands. Grepped for "getWorkingCopy()".
Maniphest Tasks: T13490, T11968
Differential Revision: https://secure.phabricator.com/D21075
Summary:
Ref T11968. Ref T13490. These are the workflows which currently use the intermediate-level hardpoint code (which made hardpoints formal, but didn't do the yield stuff).
Move toward updating them by doing some basic bookkeeping, with a few compatibility adjustments to the parent Workflow class.
Test Plan: Ran "arc branch" and "arc browse" with various arguments.
Maniphest Tasks: T13490, T11968
Differential Revision: https://secure.phabricator.com/D21074
Summary:
This reverts commit 7e25288f49. See T13507 for discussion, shortly.
This needs to be switched to capability detection because some server configurations can not actually accept these requests.
Test Plan: Straight revert.
Differential Revision: https://secure.phabricator.com/D21076
Summary: Ref T13507. Enable compression in the body of Conduit requests if we have client support for it; we expect the server should always support it.
Test Plan: Created this revision, ran random "arc" commands that use Conduit. Added debugging code, saw payload size drop.
Maniphest Tasks: T13507
Differential Revision: https://secure.phabricator.com/D21073
Summary:
Depends on D21071. Ref T11968. Currently, "ConduitEngine" tries to lightly parallelize futures. This was a compromise when the initial "hardpoint" change didn't plan to pursue real request paralleization.
Now that the newer hardpoint change does, we don't need onboard resolution in ConduitEngine. Throw it away.
When the engine is supposed to resolve a future, it now just resolves that future on its own. This should be functionally identical to the previous behavior, except that it may be slower.
(In practice, because HTTP futures are backed by an internal cURL request pool, this proably has little effect anywhere. Moving to modern hardpoints will make performance no worse than it was prior to this change, in any case.)
Test Plan: Ran various modern "arc" commands.
Maniphest Tasks: T11968
Differential Revision: https://secure.phabricator.com/D21072
Summary:
Depends on D21070. Ref T11968. Adds "yield"-aware query classes for parallelizing service calls.
These will replace the similar (but not yield-aware) "Hardpoint" classes introduced previously. This is an API change but most of the old classes still exist and still do the same thing, just with more "yield" statements.
This just adds a bunch of new code with no callers and no API updates.
Test Plan: See future changes.
Maniphest Tasks: T11968
Differential Revision: https://secure.phabricator.com/D21071
Summary:
Ref T11968.
- Allow "WorkingCopy" objects to maintain an API object and update callers ("get...()" instead of "new...()").
- Always generate a WorkingCopy object and a RepositoryAPI object.
Currently, code has to look like this:
```
$working_copy = ...
if ($working_copy) {
$repository_api = ...
if ($repository_api [instanceof ... ]) {
```
This is clunky. There's also no reason some "arc" commands can't run outside a VCS working directory without special-casing how they interact with the filesystem.
Conceptually, model the filesystem as a trivial VCS (which stores exactly one commit, always amends onto it, and discards history). Provide a trivial WorkingCopy and API for it.
(This change isn't terribly interesting on its own, but chips away at landing the new Hardpoint infrastructure.)
Test Plan: Ran `arc version`, `arc upgrade`.
Maniphest Tasks: T11968
Differential Revision: https://secure.phabricator.com/D21070
Summary:
Depends on D21067. Ref T13492. Converting unit tests to be readable exposed this error in the grammar.
Normally, "grammar_rule" rules emit a standalone Node. In this case, the bottom-level grammar rule is a collection of trivial rules and the callers configure the Node. Wrap the bottom-level rule in a configuration rule so the node is configured correctly and consistently, in exactly one place.
Test Plan: Ran unit tests.
Maniphest Tasks: T13492
Differential Revision: https://secure.phabricator.com/D21068
Summary:
Depends on D21066. Ref T13492. The switch of unit test data to stable/readable output exposed this bug in parsing of variadic calls: some nodes are not given types properly.
Fix the parser and update the test.
Test Plan: Ran the test, which now works.
Maniphest Tasks: T13492
Differential Revision: https://secure.phabricator.com/D21067
Summary:
Depends on D21065. Ref T13492. Swap existing "expect" blocks from unstable, unreadable JSON to readable, stable trees.
(There are two "INVALID TYPE" outputs which this update effectively detects and which future changes correct.)
Test Plan: Ran "arc unit --everything", got a clean build.
Maniphest Tasks: T13492
Differential Revision: https://secure.phabricator.com/D21066
Summary:
Depends on D21064. Ref T13492. Earlier, see D17819. This is essentially the same change, although I inlined the token stream into the node list.
This intentionally breaks most tests since it just has the new "expect" generator; the next change will fix them by swapping the test bodies.
Test Plan: Ran "arc unit --everything" after the next change (which fixes all the tests), got a clean build. This change on its own fails all existing XHPAST tests since the block formats don't match.
Maniphest Tasks: T13492
Differential Revision: https://secure.phabricator.com/D21065
Summary:
Depends on D21063. Ref T13492. Currently, XHPAST defines a version in both PHP code and C code, and they must be kept in sync.
Switch to a single definition in PHP, then carry it through the build pipeline into C.
Test Plan: Did a clean rebuild of XHPAST, saw a version number carried in from PHP. Ran "xhpast --version".
Maniphest Tasks: T13492
Differential Revision: https://secure.phabricator.com/D21064
Summary:
Fixes T9753. Changes some time ago (in D13970 + D13974) improved XHPAST build compile-time warning behavior under Bison 3.
However, macOS still ships with Bison 2.3 and these changes prevent XHPAST from building with Bison 2.3. The changes didn't introduce version detection, so Bison 2.3 builds fail somewhat mysteriously without obvious next steps.
It's relatively easy to install Bison 3 on macOS via Homebrew, but the Bison 3 changes aren't terribly substantive and XHPAST doesn't actually depend on any Bison 3 features, so just return to Bison 2.3 for now.
It would be reasonable to undo this again and retarget Bison 3 in the future, but ideally we should wait until macOS ships with Bison 3 or we have a specific reason to bump the minimum required version to 3. If/when we do, we should version-detect Bison and raise a clear error message.
Test Plan: Built xhpast under Bison 2.3 on a default macOS install using "make cleanall && make install".
Maniphest Tasks: T9753
Differential Revision: https://secure.phabricator.com/D21063
Summary: Ref T13490. The new Arcanist runtime supports workflow signal handling, but Phage isn't quite able to make use of it. Clean up the last few pieces so it can work.
Test Plan: Ran "phage", hit ^C, got status information.
Maniphest Tasks: T13490
Differential Revision: https://secure.phabricator.com/D21060
Summary: Ref T11968. Phage has another "sustained pool of Futures" use case, and needs some slight adjustments after Future API changes.
Test Plan: Ran `bin/phage status ...`, got a clean result instead of a JSON decoding failure.
Maniphest Tasks: T11968
Differential Revision: https://secure.phabricator.com/D21058
Summary:
Ref T11968. "FutureIterator" recently became non-rewindable, and starting a Future twice is now an error.
This complicates a handful of use cases where a mostly-constant pool of futures is maintained over a long period of time, notably in daemon overseers and repository pull daemons.
They previously relied on being able to do "new FutureIterator($futures)" to continue resolution of a list of futures from any state. This no longer works quite like it used to, since Futures generally may not belong to more than one iterator now (and this property is desirable).
Introduce "FuturePool", which maintains exactly one iterator but manages the small amount of glue around adding and removing Futures from it, destroying it if the pool empties, and rebuilding it if the pool fills.
Test Plan: See next change, which makes use of this.
Maniphest Tasks: T11968
Differential Revision: https://secure.phabricator.com/D21053
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:
Ref T13504. Ref T13209. We currently use "escapeshellarg()" under Windows, which destructively replaces "%" symbols with spaces. This is wrong, breaks general behavior with "git log --format=...", and has led to a lot of weird workarounds in `arc` code where we don't escape arguments we should be escaping to tiptoe around issues with "%".
Now that execution occurs via "bypass_shell", we can safely (probably?) escape things, although programs are still apparently free to parse the command string however they want.
Test Plan: Added unit tests, ran them under Mac and Windows, got clean results. Ran "arc version" on Mac and Windows.
Maniphest Tasks: T13504, T13209
Differential Revision: https://secure.phabricator.com/D21051
Summary:
Ref T13504. The API has changed here slightly, and if you run "arc version" without "arcanist/" being a Git working copy, it currently fatals in a misleading way.
Instead, reach the error properly.
Test Plan: Ran "arc version" after moving aside ".git/", got a helpful error message instead of a confusing "call on null" exception.
Maniphest Tasks: T13504
Differential Revision: https://secure.phabricator.com/D21050
Summary:
Fixes T13504. This fixes three issues:
# In ExecFuture, "proc_open()" on an invalid binary could fail with an unconditional exception.
# In ExecPassthru, "proc_open()" on an invalid binary could fail with an unconditional exception.
# In "arc browse", "start <url>" does not work when the shell is bypassed.
In (1) and (2), the desired behavior is to fail with an exit code which is sometimes upgraded to an exception depending on calling convention.
Issue (1) most commonly manifested as "find" failing when run via "cmd.exe".
Issue (2) most commonly manifested as "arc browse" failing.
Issue (3) was entangled with issue (2).
In cases (1) and (2), assume "proc_open()" failures under Windows are because of bad binaries and treat them like bogus commands on Linux/Mac.
In case (3), use "cmd /c start" instead of "start" as a default browser on Windows.
Test Plan:
- On Windows, did mime type detection in cmd.exe. Before patch: proc_open() exception in "find". After patch: clean (albeit not terribly useful) detection.
- On Windows, did "arc browse ...". Before patch: proc_open() exception in "start". After patch: clean browser execution.
Maniphest Tasks: T13504
Differential Revision: https://secure.phabricator.com/D21047
Summary:
Ref T11968.
- Rekeying natural lists of futures creates more problems than it solves. For now, don't try to deal with this.
- The scope of the future autokey was incorrect (per subclass, not per process).
Test Plan: Ran various future-oriented flows on new HardpointEngine code.
Maniphest Tasks: T11968
Differential Revision: https://secure.phabricator.com/D21046
Summary:
Depends on D21036. Ref T11968. Ref T13177. Currently, each Future integrates separately with ServiceProfiler, but much of the code is similar.
Move integration to the base class and lift up most of the implementation details.
Test Plan: Ran `arc diff --trace`, saw sensible output.
Maniphest Tasks: T13177, T11968
Differential Revision: https://secure.phabricator.com/D21038
Summary: Depends on D21035. Ref T11968. This allows a "FutureIterator" to hold futures which are blocked because of unresolved dependencies, and makes the resolution process more structured.
Test Plan: Ran unit tests, created this revision.
Maniphest Tasks: T11968
Differential Revision: https://secure.phabricator.com/D21036
Summary: Depends on D21033. Ref T11968. This is just an incremental step in modernizing Future and making it more robust. Currently, subclasses are expected to write directly to `$this->result`, but this isn't consistent with how modern classes generally work.
Test Plan: Ran unit tests, created this revision.
Maniphest Tasks: T11968
Differential Revision: https://secure.phabricator.com/D21034
Summary:
Depends on D21032. Ref T11968. Currently, "Future" and "FutureIterator" can both resolve futures.
Treat "Future->resolve()" as sugar on resolving an iterator of size 1.
Test Plan: Ran tests, created this revision.
Maniphest Tasks: T11968
Differential Revision: https://secure.phabricator.com/D21033
Summary:
Ref T11968. This future-level parameter has no nontrivial callers and makes the "fate of FutureGraph" changes more difficult.
Callers that are genuinely interested in this behavior can wrap the Future in a FutureIterator and use "setUpdateInterval()" to get the same behavior.
Test Plan: Grepped for "resolve()" and "resolvex()", updated callers.
Maniphest Tasks: T11968
Differential Revision: https://secure.phabricator.com/D21031
Summary:
One of our users ran into an issue where they were running `arc diff` in
a subdirectory and it so happened that in that directoy there wasn't any `lfs`
files so the repo was marked as `$is_lfs == false`. Lets allow `arc` to be run
from anywhere.
Test Plan: Internally this patch worked
Reviewers: epriestley, #blessed_reviewers
Reviewed By: epriestley, #blessed_reviewers
Subscribers: Korvin
Differential Revision: https://secure.phabricator.com/D21045
Summary:
Ref T13499. Currently, we throw most language errors as exceptions, but this list isn't exhaustive and errors we don't specifically make more severe are allowed to slip through.
This is generally undesirable, particularly in the case of "Undefined index:" errors. See T13499 for a specific case where this caused behavior to be more difficult to understand and diagnose than it should have been.
Make this the default behavior instead, except for "E_USER" errors, which we never expect to arise from first-party code.
This may be slightly too aggressive, but future changes can selectively reduce the severity of some types of errors if problems arise.
Test Plan:
- Executed code which intentionally accessed an undefined index, got an exception.
- Poked around Phabricator and Arcanist without any further issues cropping up, but I don't have a good way to develop confidence that the "reduced severity" list should genuinely be empty.
Maniphest Tasks: T13499
Differential Revision: https://secure.phabricator.com/D21044
Summary: Fixes T13498. Currently, `arc diff` may invoke the file upload flow using an older variation of the API. Both the modern and fallback clients build an Engine, so switching to the modern API is trivial.
Test Plan:
- Created a local commit with a 1MB binary file.
- Ran `arc diff --only`.
- Before patch: fatal on old API call.
- After patch: clean upload.
Maniphest Tasks: T13498
Differential Revision: https://secure.phabricator.com/D21043
Summary:
I'm working on a fairly large monorepo using phabricator and we are
running into a `git-lfs` issue. Every single `arc diff` we have to wait
many seconds while all our lfs files are walked before the push goes through.
I chatted with the `git-lfs` folks in
https://github.com/git-lfs/git-lfs/issues/4076 and they suggested this workflow
change. I tested it on our repo and it does indeed seem to fix the issue. Three
things were change:
1. The remote url is named
2. Use branches instead of tags
3. Do a `git fetch`
(Also, hello after all this time! Hope you're all doing well over at phacility.)
Test Plan:
Before
```
$ arc diff
Linting...
LINT OKAY No lint problems.
Running unit tests...
UNIT OKAY No unit test failures.
PUSH STAGING Pushing changes to staging area...
Uploading LFS objects: 100% (8211/8211), 1.0 GB | 0 B/s, done.
Enumerating objects: 31, done.
Counting objects: 100% (31/31), done.
Delta compression using up to 12 threads
Compressing objects: 100% (21/21), done.
Writing objects: 100% (24/24), 1.77 KiB | 8.00 KiB/s, done.
Total 24 (delta 19), reused 0 (delta 0)
remote: Resolving deltas: 100% (19/19), completed with 6 local objects.
To github.com:robinhoodmarkets/web-staging.git
* [new tag] 2581578b59b6341c1023377fd8741814e89fcd0c -> phabricator/base/427095
* [new tag] 7d9be50fbca590f15742d8cc20edd4f082dfbb3b -> phabricator/diff/427095
Updated an existing Differential revision:
Revision URI: https://phabricator.robinhood.com/D135442
Included changes:
M __shared/tools/arcanist-js/src/workflow/RHArcanistDiffWorkflow.php
```
After
```
$ arc diff
Linting...
LINT OKAY No lint problems.
Running unit tests...
UNIT OKAY No unit test failures.
PUSH STAGING Pushing changes to staging area...
Enumerating objects: 15, done.
Counting objects: 100% (15/15), done.
Delta compression using up to 12 threads
Compressing objects: 100% (7/7), done.
Writing objects: 100% (8/8), 638 bytes | 2.00 KiB/s, done.
Total 8 (delta 6), reused 0 (delta 0)
remote: Resolving deltas: 100% (6/6), completed with 6 local objects.
To github.com:robinhoodmarkets/web-staging.git
* [new branch] 2581578b59b6341c1023377fd8741814e89fcd0c -> phabricator/base/427105
* [new branch] 274cc3a25eb7856291c1d4c078b28b76af8a43b2 -> phabricator/diff/427105
Updated an existing Differential revision:
Revision URI: https://phabricator.robinhood.com/D135442
Included changes:
M __shared/tools/arcanist-js/src/workflow/RHArcanistDiffWorkflow.php
```
(notice the lack of `Uploading LFS objects`)
Reviewers: epriestley, #blessed_reviewers
Reviewed By: epriestley, #blessed_reviewers
Subscribers: Korvin
Differential Revision: https://secure.phabricator.com/D21041
Summary:
Depends on D21031. Ref T11968. This clears some lint with the word "Future", but is just a minor cleanup patch not really related to the main goal in that task.
We currently warn on `preg_quote()` with no second parameter, but this is valid and correct:
```
preg_match('('.preg_quote($x).')', ...)
```
This is the modern recommended style for this sort of expression, so the warning is often a false positive; drop it.
The "__CLASS__" warning looks for hard-coded class names in strings, but currently looks for them as a word anywhere in the string.
This is often a false positive and sometimes makes error messages or exceptions untranslatable. For example, translators can't do anything with this:
> Filesystem path "%s" does not exist.
...if it's written as:
> %s path "%s" does not exist.
...just because it happens to appear in the class "Filesystem". Some other classes, including "Future", suffer from this.
Even when the detection is correct, it's clunky and a mistake here (failing to update the class name in an error message) is unlikely to ever do any real damage.
Test Plan: Ran unit tests and `arc lint`.
Maniphest Tasks: T11968
Differential Revision: https://secure.phabricator.com/D21032
Summary:
See PHI1663. If "phabricator.uri" is not configured, we try to fall back to "default", but doesn't have a modern config specification and fails.
Instead, read "default" as a raw config value to preserve compatible behavior. My intent is to eventually remove it.
Test Plan:
In a directory with no "phabricator.uri" config available, ran `echo {} | arc call-conduit conduit.ping`. After the patch, got a reasonable error instead of a fatal.
In a directory with "default" configured but not "phabricator.uri", ran the same command. After the patch, got a ping.
Maniphest Tasks: T13497
Differential Revision: https://secure.phabricator.com/D21039
Summary: Ref T13490. This largely makes "arc upload" work, although the Future stuff is still a bit wonky. See T11968.
Test Plan: Ran "arc upload README.md", got an upload.
Maniphest Tasks: T13490
Differential Revision: https://secure.phabricator.com/D21030
Summary: Ref T13490. This fixes one bug in D21025 (loading relative to "arcanist/" rather than the parent directory) and improves behavior when a library fails to load (we prompt the user to continue).
Test Plan:
- Ran `arc list` with bad library specifications, got a sensible error and an option to continue.
- Ran `arc list` with a library specification that existed next to "arcanist/", got a library load instead of an error.
Maniphest Tasks: T13490
Differential Revision: https://secure.phabricator.com/D21029
Summary: Ref T13490. This makes the "phage" toolset work properly and updates external library behavior to match the "classic" behavior, except that "--library" is now supported.
Test Plan: Ran "phage remote" workflows.
Maniphest Tasks: T13490
Differential Revision: https://secure.phabricator.com/D21025
Summary:
Ref T13490. There's one simple "implode()" order issue here, and one slightly more complex one that uses "DIRECTORY_SEPARATOR" as glue.
Add test coverage for this, update the lint check to detect constants used as glue, and fix the callsites.
Test Plan:
- Added a failing test and made it pass.
- Ran `arc lint --everything` and looked for remaining implode warnings, found none.
Maniphest Tasks: T13490
Differential Revision: https://secure.phabricator.com/D21024
Summary: Ref T13490. Allows "arc upgrade" to run through the new Toolsets infrastructure.
Test Plan: Ran "arc upgrade", although you can't upgrade feature branches so this can't be entirely tested until it hits "master".
Maniphest Tasks: T13490
Differential Revision: https://secure.phabricator.com/D21006
Summary: Depends on D21004. Ref T13490. The "wilds" branch removed a bunch of dead features but some of them were resurrected by the merge. I don't think any of these are desirable to retain, so purge them all again, even in classic mode.
Test Plan: Grepped for affected symbols.
Maniphest Tasks: T13490
Differential Revision: https://secure.phabricator.com/D21005
Summary:
Ref T13490. Historically, "arc" was eager to load external libraries from many locations. I plan to make this significantly more structured/safer in the future (see T5055), but this rule got dropped as the "experimental" and "wilds" branches collapsed to "master".
Restore "next to arcanist" and "inside externals/includes" as valid search locations for classic workflows.
Test Plan: Ran "arc" commands in a working copy with a "load" library configured.
Maniphest Tasks: T13490
Differential Revision: https://secure.phabricator.com/D21004
Summary:
Depends on D21001. Ref T13490.
- Require "--" to terminate arguments when running noninteractively.
- Don't correct spelling when running noninteractively (we still suggest corrections).
- Remove old "phage" wrapper script.
- Fix an issue where "argv" could implicitly generate a parseable "--argv" flag.
- Accept "--" as an argument terminator even when there is no argument list.
- Update some strings to use more modern quoting style.
- Make workflows decline to handle signals by default.
- Allow "arc weld" to weld non-UTF8 files together.
Test Plan: Ran various commands. Welded non-UTF8 files.
Maniphest Tasks: T13490
Differential Revision: https://secure.phabricator.com/D21002
Summary: Ref T13490. Update the "weld" workflow and the "anoid" workflow. Incorporates D20938.
Test Plan: Ran "arc weld". Ran "arc anoid".
Maniphest Tasks: T13490
Differential Revision: https://secure.phabricator.com/D21001
Summary: Depends on D20996. Ref T13395. Ports the updated version of this workflow from "experimental".
Test Plan: Ran `arc shell-complete` to install the hook, then shell-completed commands.
Maniphest Tasks: T13395
Differential Revision: https://secure.phabricator.com/D20997
Summary: Depends on D20993. Ref T13395. Merges the more-modern "alias" out of "experimental".
Test Plan: Defined and invoked aliases. This has some rough edges: notably, no easy list/delete flow.
Maniphest Tasks: T13395
Differential Revision: https://secure.phabricator.com/D20996
Summary: Ref T13395. Depends on D20991. Make "arc liberate" run as a toolset command, not a classic command.
Test Plan: Ran "arc liberate"; created this diff.
Maniphest Tasks: T13395
Differential Revision: https://secure.phabricator.com/D20992
Summary: Ref T13395. Make "arc help" run as a toolset command, not a classic command.
Test Plan: Ran "arc help".
Maniphest Tasks: T13395
Differential Revision: https://secure.phabricator.com/D20991
Summary:
Depends on D20988. Ref T13395. Ref T13098. Ref T13203.
This brings all the "toolsets" code into "master". We try to run commands as toolsets commands first, then fall back to older code.
Since the "toolsets" class tree is mostly parallel to the older class tree, this isn't completely broken. Currently, all commands fall back.
Test Plan: Created this diff, ran various other commands. But this is probably a long shot from finished.
Maniphest Tasks: T13395, T13203, T13098
Differential Revision: https://secure.phabricator.com/D20990
