Re: Round-tripping fast-export/import changes commit hashes

On Tue, Aug 10, 2021 at 8:51 AM anatoly techtonik [off-list ref] wrote:

On Mon, Aug 9, 2021 at 9:15 PM Elijah Newren [off-list ref] wrote:

quoted

The author wasn't interested in implementing that
suggestion (and it's a low priority for me that I may never get around
to).  The series also wasn't pushed through and eventually was
dropped.

What it takes to validate the commit signature?

I'm not familiar with any of the gpg libraries, and don't even have an
active gpg key.  So, I don't know.  Some quick grepping shows that we
have gpg-interface.[ch], so we have some functions we can apparently
call.

Isn't it the same as validating commit tag?

gpg signatures of tags are somewhat different than gpg signatures of commits:

* gpg signatures for tags are simply part of the annotated tag message
* gpg signatures for commits are stored in a separate commit header,
not just as extra text at the end of the commit message

This gpg signature handling for tags means that fast-import isn't even
aware of whether the tag is signed; it simply sees a commit message
and records it.  fast-export also would have been unaware and just
exported them as-is if someone hadn't written some special parsing for
it.  fast-import would need to do similar special parsing to become
aware of whether the tags are signed or not.  For now, fast-import
just keeps any tag messages as-is, and thus potentially writes invalid
tag signatures.  (The only way people have to control this is at the
fast-export side with the --signed-tags flag, which gives you the
choices of abort, strip, or keep the signatures even though they'll
likely be wrong.)  If fast-import were to gain knowledge of tag
signatures and an ability to validate them, it could offer smarter
options like keep-if-valid-and-discard-otherwise.

In contrast, the fact that gpg signatures for commits have to be
recorded as a separate commit header means they cannot be recorded in
fast-import without additional code changes.  And both the fast-export
and fast-import sides have to be made aware of and specially handle
the commit signatures for them to even get propagated, let alone
validated.

Is it possible to merge at least the `--fast-export`
part? The effect of roundtrip would be the same, but at least external
tools would be able to detect signed commits and warn users.

The fact that it wasn't merged suggests there was some issue raised in
feedback that wasn't addressed.  I don't remember if that was the case
or not, but someone would have to find out, address any remaining
issues pointed out by feedback, and champion it through.

Personally, I don't like shoving a half solution through and think
there needs to be validation on the fast-import side added at the same
time, but others may disagree with me.  I have plenty of other
projects to work on, though, so whoever does the work will more likely
be the ones to decide.

quoted

[1] https://lore.kernel.org/git/20210430232537.1131641-1-lukeshu@lukeshu.com/ (local)

quoted

Yes, and I mentioned several other reasons why a round-trip from
fast-export through fast-import cannot be relied upon to preserve
object hashes.

Yes, I understand that. What would be the recommended way to detect
which commits would change as a result of the round-trip? It will then
be possible to warn users in `reposurgeon` `lint` command.

There is no function or command that would check that kind of thing
short of doing the round-trip.  I provided a list of reasons IDs could
change as a starting point in case anyone wanted to try to write a
function or command that could check, and to point out that it is a
long list and might grow in the future.

I think practically, if you're doing a one-shot export (as I
originally assumed from your email), that you'd find out and then just
manually fix things up by hand.  If your goal is writing or changing a
general purpose filtering tool, then I'd suggest instead using the
alternate technique I outlined in the other thread you started at [2].

[2] https://lore.kernel.org/git/CABPp-BH4dcsW52immJpTjgY5LjaVfKrY9MaUOnKT3byi2tBPpg@mail.gmail.com/ (local)

quoted

(3) fast-export works by looking for the relevant bits it knows how to
export.  You'd have to redesign it to fully parse every bit of data in
each object it looks at, throw errors if it didn't recognize any, and
make sure it exports all the bits.  That might be difficult since it's
hard to know how to future proof it.  How do you guarantee you've
printed every field in a commit struct, when that struct might gain
new fields in the future?  (This is especially challenging since
fast-export/fast-import might not be considered core tools, or at
least don't get as much attention as the "truly core" parts of git;
see https://lore.kernel.org/git/xmqq36mxdnpz.fsf@gitster-ct.c.googlers.com/ (local))

Looks like the only way to make it forward compatible is to introduce
some kind of versioning and a validation schema like protobuf. Otherwise
writing an importer and exporter for each and every thing that may
encounter in a git stream may be unrealistic, yes.

quoted

quoted

P.S. I am resurrecting the old thread, because my problem with editing
the history of the repository with an external tool still can not be solved.

Sure it can, just use fast-export's --reference-excluded-parents
option and don't export commits you know you won't need to change.

How does `--reference-excluded-parents` help to read signed commits?

It doesn't.  I was assuming you were doing a one shot export, namely
of the repository you linked to,
https://github.com/simons-public/protonfixes, and that you already
knew which commits were not going to be changed (because you pointed
them out in your email to the list) -- and in fact that it was only a
single commit affected, as you mentioned.

Armed with that knowledge, you could just export the parts of the
repository AFTER that commit, and use --reference-excluded-parents to
make sure the fast-export stream built upon them rather than squashing
all changes up to that point into the first commit in the stream.

`reposurgeon` needs all commits to select those that are needed by
different criteria. It is hard to tell which commits are not important without
reading and processing them first.

Right, so you aren't trying to just handle this one repository, but
modify/create a general purpose tool that does so.  See my response in
the other thread you started, again at [2] above.

quoted

Or, if for some reason you are really set on exporting everything and
then editing, then go ahead and create the full fast-export output,
including with all your edits, and then post-process it manually
before feeding to fast-import.  In particular, in the post-processing
step find the commits that were problematic that you know won't be
modified, such as your signed commit.  Then go edit that fast-export
dump and (a) remove the dump of the no-longer-signed signed commit
(because you don't want it), and (b) replace any references to the
no-longer-signed-commit (e.g. "from :12") to instead use the hash of
the actual original signed commit (e.g. "from
d3d24b63446c7d06586eaa51764ff0c619113f09").  If you do that, then git
fast-import will just build the new commits on the existing signed
commit instead of on some new commit that is missing the signature.
Technically, you can even skip step (a), as all it will do is produce
an extra commit in your repository that isn't used and thus will be
garbage collected later.

The problem is to detect problematic signed commits, because as I
understand `fast-export` doesn't give any signs if commits were signed
before the export.

Signed commits is just one issue, and you'll have to add special code
to handle a bunch of other special cases if you go down this route.
I'd rephrase the problem.  You want to know when _your tool_ (e.g.
reposurgeon since you refer to it multiple times; I'm guessing you're
contributing to it?) has not modified a commit or any of its
ancestors, and when it hasn't, then _your tool_ should remove that
commit from the fast-export stream and replace any references to it by
the original commit's object id.  I outlined how to do this in [2],
referenced above, making use of the --show-original-ids flag to
fast-export.  If you do that, then for any commits which you haven't
modified (including not modifying any of its ancestors), then you'll
keep the same commits as-is with no stripping of gpg-signatures or
canonicalization of objects, so that you'll have the exact same commit
IDs.  Further, you can do this today, without any changes to git
fast-export or git fast-import.