Use pycompat.fsencode

Just saw yuya's comment, I will do that.

Add to whitelist

Add to whilelist

Use a different function for getting hexdigest as bytes

In D5056#75750, @pulkit wrote:

I took care of this in https://phab.mercurial-scm.org/D5040 ;)

Add whitelist for test this fixes.

Rebase and fix conflicts

This is necessary to recreate something like the original output for hg absorb using the template language. It also allows inclusion of the commit message in the rows of the absorb output, using a template spec like:

hg absorb -pn -T '{ifeq(linetype,"summary","\n{count} changesets affected",ifeq(linetype,"path","showing changes for {label("absorb.path",path)}","{label("absorb.node", pad(node, 7, truncate=true))} {label("yellow", pad(desc, 30, truncate=true))} {label("diff.{linetype}", "{diffchar}{inserted}{deleted}{hunk}")}"))}\n'

I'm willing to entertain better suggestions, but until Cargo has proper support for vendoring git repositories, we're stuck with an ugly hack.

Looks reasonable to me. This could also be achieved by making SSH a script which sets the environment up and invokes the real SSH, but that's clumsy, whereas this allows use of config to set the env vars.

I don't think it's possible to reliably provoke the bad behaviour. What's happening here is we are using the bound identity method of the dirstatemap instance as the identity of the dirstate. This nearly always changes when the dirstate is recreated (which happens when the dirstate file changes on disk by virtue of dirstate being a filecache property and by localrepo.invalidatedirstate. The only time this will actually cause problems is if the new dirstatemap object ends up with the same address as the old one. In that case we will think the dirstate hasn't changed (as bound methods are compared using the combination of object address and method address), and potentially update using an old dirstate instance.

I'm not actually sure what problem it would cause, I just know it's wrong. I think the only place we actually use identity is in deciding when to update the dirstate after a status call that discovers the dirstate isn't right (e.g. normallookup files can now be marked normal). Normally status doesn't have the lock, so it optimistically grabs the lock, but only does the update if the identity matches. I don't really have a good idea what the function would do when it's using the bound method rather than the real identity. Suggestions for a test for this race?