Updating the commit message for reverts

Hi,

I work on the Gerrit team and I would like to change the default behavior for suggested commit messages for Reverts. Currently, if the user is trying a change called '"Revert "change X"', the suggested commit message would be 'Revert "Revert "Change X""' which is silly, since sometimes users want to revert the same change many times.

The suggestion is to change the behavior to "Revert^N" instead of multiple Reverts one after another.

I'm happy to change those things in Gerrit, but it would also be nice if it were changed inside of Git.

What do you think?

Thanks! Gal Paikin

Gal Paikin <paiking@google.com> writes:

I do not _think_ anybody lets the exact phrase 'Revert ...' in the log message to be read by scripts to perform machannical action, so in that sense it may be OK to special-case the reversion of a revert and rephrase it in a more human friendly way.

BUT

 * what does "Revert^47" even mean?  Not just the proposed phrasing
   looks horrible, it is not even clear what happened at the end.
   Was the patch turned out to be OK after all these reversion war,
   or got rejected for now?  It also misleads readers who know Git
   can perform a merge with more than two parents that it may be
   reverting the effect relative to 47th parent of the commit.

   It _might_ be slightly more acceptable to flip the phrase between
   "Revert X" and "Reinstate X" (or "Reapply X"), without saying
   "this is the 47th round of our reversion war".  I dunno.

 * how often does it happen in practice?  If a group of developers
   find themselves reverting and reapplying the same commit more
   than a few times, wouldn't they rather stop and think before
   doing yet another round, which I expect to result in a better fix
   implemented as a separate and brand new patch that takes
   inspiration from a patch that was earlier reverted, and at that
   point it won't be the 47th iteration of reversion war anyway.

So, I am fairly negative on the change in the proposed form as-is.

Hi, Thanks for the reply!

So the idea of changing from "Revert Revert" to "Reland", "reapply" has a big problem: sometimes Revert^2 actually means 'reverting "Revert"' since "Revert" introduced a bug that wasn't in the original change.

So to your question, I don't know what Revert^47 means since it depends on each individual case. Sometimes it actually means "Revert" and sometimes it means "Reland".

So do people actually use it? Yes! Many users reported to me that it is not that unusual to get to "Revert^6", and it is very usual and common to get to "Revert^2/3/4". It is also useful for the users to know the number of the revert, according to the reports. Here is an example: https://android-review.googlesource.com/c/platform/art/+/352330 Feel free to also search for "Revert^2/3/4" to find many results.

Anyway, I am certain that "Revert^3" is better than "Revert revert revert". There is definitely no clear way to solve this issue, but perhaps "nth revert" would be a more "human language" solution?

Best, Gal

On Tue, Dec 24, 2019 at 8:15 PM Junio C Hamano <gitster@pobox.com> wrote:

On 2019-12-27 11:13:47+0100, Gal Paikin <paiking@google.com> wrote:

I've seen Revert x6 in a code base, I couldn't get to know the reason for that reversion war. I think it could be seen more in some in-house web development that uses trunk-based development, code is being tested with CI/CD, lightly tested, squash-merged to master, then run into problem in staging (or worst, production, because not enough traffix was generated for testing environment).

In my very personal opinion, "nth revert" is a poor choice. At a first glance, I would take it as:

	This is the "nth revert", after applying this patch n times.

Gal Paikin <paiking@google.com> writes:

Sorry, I do not see a relevance of the above in this discussion, as the situation does not improve if you phrase it as "Revert^2" or "Second Revert".

Also as somebody else said in downthread, the phrasing "second revert" would typically mean "a patch gets applied, proves to be premature and gets reverted, the revert is reverted because the situation in the rest of the system improved to make the orignal patch viable, and then gets reverted again due to some other issues", i.e. "Revert Revert Revert do something", so it is even worse.

(also cc-ing the Gerrit list. The git mailing list requires plain-text
 email and prefers bottom-posting; there shouldn't be any other
 gotchas if you reply-all)
Hi Gal,

Gal Paikin wrote:

Thanks for writing. I like it when tools behave consistently with one another, so I'm glad you brought it up here.

For context, [1] is where this was discussed previously on the Gerrit mailing list. Do you have more context? For example, have there been reports from users requesting this kind of change?

For a bit of precedent, the Chromium project has some custom logic[2] that rewrites a revert of a revert to "Reland". I kind of like that, since it makes the intention behind the revert-of-revert a little clearer.

If both reverts are clean, then the typical use case I can think of is that a prerequisite was missing, hence the revert; after all the prerequisites were in place, it was then safe to roll forward, hence the reland. I would hope that the full commit messages for these revert commits would provide this context since without that kind of communication from the people involved, a reader is lost.

This would be replacing one kind of jargon with another, so it's not clear to me that it would improve matters.

With Revert / Reland, we can forget the N altogether:

 1. 'Do some great thing'
 2. 'Revert "Do some great thing"'
 3. 'Reland "Do some great thing"'
 4. 'Revert "Do some great thing"'
 5. (etc)

For the reader of the shortlog, it's not too important how long the edit war has gone. The single word makes it clear what the commit is going to do.

In the full commit message, I would hope that people put in some explanation of the story so far to explain why they have gone back-and-forth all these times.

Thanks and hope that helps, Jonathan

[1] https://groups.google.com/d/msg/repo-discuss/edl43LcyEKE/f9Q65kOcBgAJ [2] https://chromium-review.googlesource.com/c/infra/gerrit-plugins/chromium-behavior/+/757684

(not cc-ing the Gerrit list after all, since only members can post) Hi Gal,

Gal Paikin wrote:

Thanks for writing. I like it when tools behave consistently with one another, so I'm glad you brought it up here.

For context, [1] is where this was discussed previously on the Gerrit mailing list. Do you have more context? For example, have there been reports from users requesting this kind of change?

For a bit of precedent, the Chromium project has some custom logic[2] that rewrites a revert of a revert to "Reland". I kind of like that, since it makes the intention behind the revert-of-revert a little clearer.

If both reverts are clean, then the typical use case I can think of is that a prerequisite was missing, hence the revert; after all the prerequisites were in place, it was then safe to roll forward, hence the reland. I would hope that the full commit messages for these revert commits would provide this context since without that kind of communication from the people involved, a reader is lost.

This would be replacing one kind of jargon with another, so it's not clear to me that it would improve matters.

With Revert / Reland, we can forget the N altogether:

 1. 'Do some great thing'
 2. 'Revert "Do some great thing"'
 3. 'Reland "Do some great thing"'
 4. 'Revert "Do some great thing"'
 5. (etc)

For the reader of the shortlog, it's not too important how long the edit war has gone. The single word makes it clear what the commit is going to do.

In the full commit message, I would hope that people put in some explanation of the story so far to explain why they have gone back-and-forth all these times.

Thanks and hope that helps, Jonathan

[1] https://groups.google.com/d/msg/repo-discuss/edl43LcyEKE/f9Q65kOcBgAJ [2] https://chromium-review.googlesource.com/c/infra/gerrit-plugins/chromium-behavior/+/757684

On Mon, Dec 30, 2019 at 11:59:15AM -0800, Jonathan Nieder wrote:

the irony here is that "reland" is of course yet more jargon. :-D i'd strongly suggest to use something that actually appears in standard dictionaries, preferentially "reapply".

in principle i agree, but it irks me somewhat that the summaries become non-unique, as that always somewhat impacts history browsing.