On 26/09/2023 18:04, Thiago Jung Bauermann wrote:
Hello Guinevere,
Guinevere Larsen blarsen@redhat.com writes:
I got 3 of these messages about a patch I have upstream, which seems excessive. I also just managed to grab an aarch64 box and test it and got no errors, so I'd appreciate someone double checking that those werent 3 false positives please.
I'm sorry for the false positive and the extra messages (Could you forward them (with full headers) to me, please? Perhaps we can prevent sending identical messages in the case of an unexpected mishap).
Yes of course! I'm not very knowledgeable in email, did you get all headers with this forward, or is it better if I attach them to a private email to you?
And thank you for testing on an aarch64 box that the patches are indeed ok.
What happened here is that the In-Reply-To: fields of the v2 messages all point to the v1 patch. IIUC, normally the v2 cover letter would point to the v1 patch, and patches 1 and 2 would point to the v2 cover letter. This confused patchwork's series detection and so our CI applied the v2 2/2 patch on top of the v1 patch instead of the v2 1/2 patch. This of course caused a build failure.
Yeah, I don't understand what happened with my git, but lucky that it did bring up the multiple replies thing at least.
I wonder if that is related? Could the CI have looked at all emails, found the parent which is a real patch, and applied 2/2 for all emails, regardless of which email was being tested due to some patch series heuristic?
Not sure how to avoid this problem in the future. Perhaps we could add an heuristic to detect this case and skip precommit testing? The heuristic could be something like looking for a "[v<n> x/y]" prefix and checking whether it is consistent among all patches in the series.
Thank you for bring this to our attention!
No problem! Thank you for the effort of improving GDB!