Perhaps too little too late: changing to ((?<!\w)\w{3,}://) also resolves the issue for me. Limiting the protocol to 32 characters seems sane either way.

Perhaps \w{3,}+ resolves the issue? (+ being a possessive quantifier here that prevents backtracking)

Thanks - I thought it was already filed but couldn't seem to find it.

I'd love folder shortcuts (for directories that only contain a single directory). Clicking through a few levels in various Java projects would be slightly more pleasant.

I understand why the behavior has changed, and I'm not particularly opposed to keeping it as-is. I can easily reach over the line number - it's just that it initially appeared as a regression so I had to note it here.

Is there anything I can do to help this move forward?

On mobile I very quickly get an inline comment box when scrolling which then takes focus so scrolling is aborted.

In D17819#215382, @richardvanvelzen wrote:

This looks a lot better already.

This isn't a completely exhaustive representation of the tree (parent nodes could still have the wrong token range and be missed by the tests) but I think it's a reasonable balance of readability and stability.

One possible solution would be to include all tokens (and their types) in the tree as well? E.g.:

<snip>

Granted, this does not make it particularly more readable, but it does solve testing token ranges for nodes.

This looks a lot better already.

Because the token values have shifted pretty much all test cases have bad data :(

Very nice

Sure, no problem. Let me know if there is something I can do to help

This does not appear to have any impact on arc/lint.

For real now :(

Remove copy-flag from new unit test (caused by copy-via-IDE)

No worries, totally understandable.

See https://secure.phabricator.com/T10382. I've mitigated by downgrading to 3.4.2 (some other version will probably work too, this just was the first one I tried)

For reference, I use this small script to allow myself to easily use arc land with bookmarks:

I'm about to start implement full dereferencing support. Cases that are missing now:

($expr)[0];
SomeClass::BRILLIANT_CONSTANT[5];

Not really high-prio, but a minor annoyance after getting a fresh xhpast version.

This is starting to bite us now - especially the coalesce construct is so useful that people started using it, taking lint syntax errors for granted.

This is almost definitely T8298, because % can't be properly escaped on Windows right now.

(I'll be away from my machine for the weekend, when accepted it'd be best if you land this since I probably won't be able to)

I'll create a dedicated PHP 5 vm to use for Phabricator development from now on. I'm not wanting to risk this kind of breakage.

The failed build is due to sort stability having been changed between 5 and 7. I'll send a diff to revert this, since phabricator doesn't support 7 officially. (T9640)

Sorry, I wasn't exhaustive in my tests. It turns out this fixed behaviour for PHP 7, but broke it on 5. I'm investigating the difference.

Remove useless %token declaration for T_COALESCE

On a sidenote, arc unit --everything in arcanist seems totally okay with these changes as well.

Was xhpast rebuilt properly? I can't seem to reproduce that failure locally.

Awesome! I'll follow up with some other features next week. Those should be a lot less tricky.

Add test case

Add variadic parameter support

In D15678#183349, @epriestley wrote:

Possibly we can squeeze ... into the same position as &, since the two are presumably exclusive? Not sure if that's desirable or ...&$x is valid.

I think the only other noteworthy change in PHP 7 that needs an additional node is variadic arguments:

Add return types in the AST

We've moved to fully utilizing PHP 7 features, and specifically miss this (regular scalar type hints are covered by normal type hints already)

One particular issue is that there is no room in the AST to add the return types. I've added TODOs at the relevant locations.

@epriestley I can add <script language="PHP">... support to XHPAST if you'd like, but since it's removed from PHP7 and never was really mainstream I don't believe it's worth the effort.

You can probably use bin/policy unlock to open up the policy.

In the extreme case where you have, say, a million tests, I'd expect there is probably little value in reporting each test into Harbormaster as a "pass", and your harness might want to summarize all passes into "999,998 additional tests passed" and submit two failures and one aggregate-pass.

Visit /config/module/versions on your install. If it does not exist, you're not running HEAD.

• You can obtain the versions at /config/module/versions/ on your install
• The max nesting level is set to PHP_INT_MAX (see libphutil/scripts/__init_script__.php:44), so you should never see this error.

FWIW, I've pretty much never looked at "Open Revisions Affecting These Files", and only look at "Local Commits" to check if the parent commit is already pushed.

In D10337#178632, @epriestley wrote:

(The n_CATCH_LIST was previously always-nonempty, right? … )

For lists, I generally see the "n_X_LIST with no members" approach.

Make sure to not skip staging (sorry about that)

Implement the restriction via the grammar