1377081 - Pruning an fxtree-labeled head breaks `fxheads` and thereby `hg wip` (fxheads() doesn't seem to care if revisions are hidden)

Status: RESOLVED FIXED

(Developer Services :: Mercurial: firefoxtree, defect)

Description

[:Gijs (he/him)]

STR:

1. pull mozilla-unified with firefoxtree enabled
2. pull autoland manually because you need to rebase something so it can land
3. prune all pulled autoland revisions with something like 'hg prune -r 409717::.'
4. run 'hg wip'

ER:
tree output as usual

AR:
abort: unknown revision 'ed1964e4da87a8d44b81c3e8ce2e88da8309af79'!

Similarly:

hg log -r 'fxheads()'

also returns this error.

Comment 1

[:Gijs (he/him)]

FWIW, just re-pulling mozilla-unified doesn't help.

Comment 2

[:Gijs (he/him)]

Workaround:

rm .hg/firefoxtrees
hg pull


(Or, presumably, selectively removing the cached head ref in there.)

But I had to go spelunking into the code for this, and it seems the docs are out of date because they still say that the code uses tags (which would be manipulable with 'normal' hg commands), which it isn't.

Comment 3

[Gregory Szorc [:gps]]

Attachment: firefoxtree: add test for unknown revision (bug 1377081);

Currently, we don't filter entries in the firefoxtrees file against
repository contents. As a result, we may inject a reference to a
revision that doesn't exist or is filtered. This will cause a
run-time error as soon as we attempt to resolve the node referenced
by a namespace entry.

Add tests for this.

Review commit: https://reviewboard.mozilla.org/r/157544/diff/#index_header
See other reviews: https://reviewboard.mozilla.org/r/157544/

Comment 4

[Gregory Szorc [:gps]]

Attachment: firefoxtree: ignore unknown revisions (bug 1377081);

As the previous tests demonstrated, we currently abort if a
reference to an unknown node leaks into the namespace. To
prevent this, we filter entries in the .hg/firefoxtrees
file at read time against known nodes.

There are side-effects to this change. For example, we'll silently
discard unknown nodes on next write. I think this is acceptable.
I'm trying to think of a scenario where silently discarding
references to old nodes is a bad thing. If a node is removed
(or hidden), we should drop our reference to it. Worst case,
if the node is re-introduced, the `hg pull` may re-populate the
firefoxtrees file if that node is still a head.

Review commit: https://reviewboard.mozilla.org/r/157546/diff/#index_header
See other reviews: https://reviewboard.mozilla.org/r/157546/

Comment 5

[:glob ✱]

Comment on attachment 8886742 [details]
firefoxtree: add test for unknown revision (bug 1377081);

https://reviewboard.mozilla.org/r/157544/#review162748

Comment 6

[:glob ✱]

Comment on attachment 8886743 [details]
firefoxtree: ignore unknown revisions (bug 1377081);

https://reviewboard.mozilla.org/r/157546/#review162752

lgtm

Comment 7

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/hgcustom/version-control-tools/rev/d2c8fc483da2
firefoxtree: add test for unknown revision ; r=glob
https://hg.mozilla.org/hgcustom/version-control-tools/rev/49edbf24215b
firefoxtree: ignore unknown revisions ; r=glob