What blocks that?

Why not @util.propertycache and do away with self._db_handle?

Or maybe pygit2 takes care of locking while it updates? So I wonder if this is fine the way it is. No action required.

nit: drop the leading \n and teach people to include newline at EOF instead?

Did you intend to call the file requires and not need this-is-git? I think this extension should also register with featuresetupfuncs.

Could we also get __iter__? We can of course add that later, but maybe it seems easy to add anyway (revlog.py has return iter(pycompat.xrange(len(self)))).

Maybe also copy the following from revlog.py?

def tiprev(self):
    return len(self.index) - 1 # well, use "len(self)" here, I guess

def tip(self):
    return self.node(self.tiprev())

def revs(self, start=0, stop=None):
    """iterate over all rev in this revlog (from start to stop)"""
    return storageutil.iterrevs(len(self), start=start, stop=stop)

Will the ? be replaced by abc123% or b'abc123%' on py3? (Same applies further down.)

I've added a TODO.md that documents what needs to happen here (missing an interface.)

Noted this in the TODO.md

This was intentional: I don't want to take a valid-but-missing-trailing-newline .git/info/exclude and blindly write a .hg at the end of an existing line. Are you saying the paranoia feels misguided?

It honestly didn't occur to me to use the presence of git in requires to trigger the "this is a git repo" behavior. Should I add a TODO about that?

I'm hoping we can only implement iter on changelog, not on baselog. Ditto for tip and revs, but I was going to block that on the interface definition (I haven't yet seen anything that wants these methods...)

It should be the former. I've actually been developing this extension exclusively on Python 3, so the tests already pass on 3.

That seems like a good idea, because there's the capability to load extensions automatically based on entries in requires. That would make it easier to not have to configure it globally.

Please do, because that seems like the natural way for it to work.

(I haven't yet seen anything that wants these methods...)

I found some things while playing with the extension :) I don't remember anymore what those were.

I think I asked because I noticed that the shortest() template function always returned the full hash and things were really slow (I assume that was because of the byte string on line 173, not here). Does it work for you? (I'm fine with adding a TODO about fixing it if you see the same brokenness.)

equivalent to return pycompat.sysstr(hex(n)) (would be using utf-8 instead of ascii, but that shouldn't matter)?

Call it gittreemanifest in that case?

I'm seeing shortest() actually returning short hashes (4-bytes, like I've got configured) so I'm not sure what's broken.

I added a test, which passes under python3.

That test case fails on py3 for me. I'll wrap the id + b'%' in pycompat.sysstr() in flight to make the test case pass.