In D12603#193005, @martinvonz wrote:

I agree, except that I've seen it crash sometimes. I wrote a little script to find such cases now. Here's an example: hg diff --change 7c2921a6. So we need to fix that first. Sorry I never bothered to report a bug for it.

In D11757#180388, @Alphare wrote:

In D11757#180387, @martinvonz wrote:

Until this is done (hopefully this cycle), we're keeping Python 2.

To make sure I understand correctly, "this cycle" here refers to the cycle *after* 6.0? So we're instead hoping that 6.1 will be the last release with support for Python 2?

This patch basically mean that if we fix thg packaging before 6.1 is released, we can drop Python 2. For us developers, that means still supporting Python 2 until this is done. Could be tomorrow (fingers crossed heh), at the end of the cycle, or in Mercurial 7.8 if nothing moves on. As said previously we're trying to make that work currently internally at Octobus, but it's really not something that we have time (or really expertise) for right now unfortunately.

Seems like a good fix, needs rebased

Seems fine, needs rebased.

In D11669#179233, @martinvonz wrote:

Sure, I could send a patch for adding another line to the help text. Do you think that's clear enough? I don't know where else it would go. (We would probably still want to override it internally to highlight it when we roll it out to Google devs.)

In D11669#179228, @martinvonz wrote:

In D11669#179227, @durin42 wrote:

Doing cross review: is there any indication of the direction of history when the UI pops up, or does the user just have to figure it out?

The user will have to figure it out (hopefully they remembered if they set the config option). Internally at Google, I was going to override helplines() (and some other function which I've forgotten now) to make it print a red banner saying if the direction has changed.

In D10893#174783, @marmoute wrote:

I have not forgotten and I meant to get to that last week. Unfortunately some IRL situation came up and I have not be able to spend time on this (and even writing this message took a week longer that I wanted). Sorry for the delay. The second half of ext week should be more "normal" and I hope to get to that at that time.

Doing cross review: is there any indication of the direction of history when the UI pops up, or does the user just have to figure it out?

In D10538#179050, @marmoute wrote:

In D10538#179035, @spectral wrote:

I'm still struggling to identify where my proposed fix makes any of this worse.
Current state: people with a version of clang-format <11, or people with a 'trunk' version of clang-format, don't get their code run through clang-format at all, and send changes that will break CI when they land (unless the issue is caught during review).
Proposed state: that no longer happens for people running a 'trunk' version of clang-format, just those using clang-format <11.

I feel like the proposal is going in the wrong direction because it make a more diverse set of version passe the version checks while I feel like we need to more toward a narrower check. Different formater version tends to produce different results and it seems saner to pin the project to specific versions (that we update from time to time).
For the record, right now we have :

• black: check says ">= 20.8b1", CI uses "20.8b1"
• clang: check says ">= 11", CI uses debian's clang-format-11 package (in practice "Debian clang-format version 11.1.0-++20210622113218+1fdec59bffc1-1~exp1~20210622213839.163")
• rust: check says "nightly-2020-10-04", CI uses "nightly-2020-10-04"

Is my feedback clearer and does it make sense to you? Or do you feel like I missed something?

I think that's a buggy test? @martinvonz and @spectral might have an opinion.