Finalizing the v4l-utils conversion to meson
Hello,
This mail (and hopefully mail thread) aims to prepare for the Media Summit 2022 discussion about finalizing the conversion of v4l-utils to meson.
The original port of v4l-utils to meson was done by Ariel D'Alessandro (big thanks for that !) and posted to the linux-media mailing list in April 2022 ([1]). Another RFC version followed ([2]), and the series then graduated to non-RFC ([3]) with new versions following ([4], [5] and [6]) until v5 ([7]) in May 2021. I believe it is time to decide if we want to move to a more modern build system or stay for the foreseable future in the past (this statement should indicate my opinion on the subject :-)).
I maintain a branch with the meson integration that I keep rebasing on top of the v4l-utils master branch. You can find it at
git://linuxtv.org/pinchartl/v4l-utils.git meson
I have also just posted the latest version of the integration patches in a v6 ([8]).
I have been using the meson integration for about 2 years now, and it has provided me with a much smoother experience than autoconf, both for native builds and cross builds. Long gone are the day where I had to fight autoconf and hack various Makefile.am to comment out pieces of the tree that would fail to compile properly and wouldn't want to get disabled through autoconf. These issues are most likely due to shortcomings in the autoconf usage in v4l-utils than problems with autoconf and automake themselves, but I quickly gave up on trying to fix that as meson just worked out of the box as intended.
This being said, I won't pretend that the current implementation would work perfectly for everybody. I twould thus like to get feedback on how to move forward.
1. Is there a general agreement that replacing autoconf is a good idea, provided that any technical issue in the proposed meson implementation (if any) can be fixed ? Or would it require fighting ophidiophobia and other non-technical issues that would make it a lost battle from the start ?
2. What are the technical issues that still need to be solved (if any) to replace autoconf with meson ?
There's no need to wait for the media summit to start answering those questions, if we can resolve the issue before meeting face to face, we'll have more time to discuss other questions :-)
[1] https://lore.kernel.org/linux-media/[email protected]
[2] https://lore.kernel.org/linux-media/[email protected]
[3] https://lore.kernel.org/linux-media/[email protected]
[4] https://lore.kernel.org/linux-media/[email protected]
[5] https://lore.kernel.org/linux-media/[email protected]
[6] https://lore.kernel.org/linux-media/[email protected]
[7] https://lore.kernel.org/linux-media/[email protected]
[8] https://lore.kernel.org/linux-media/[email protected]
Signature · 3 lines
Regards, Laurent Pinchart
Hi Laurent,
Quoted 37 lines · Laurent Pinchart
Hello,
This mail (and hopefully mail thread) aims to prepare for the Media Summit 2022 discussion about finalizing the conversion of v4l-utils to meson.
The original port of v4l-utils to meson was done by Ariel D'Alessandro (big thanks for that !) and posted to the linux-media mailing list in April 2022 ([1]). Another RFC version followed ([2]), and the series then graduated to non-RFC ([3]) with new versions following ([4], [5] and [6]) until v5 ([7]) in May 2021. I believe it is time to decide if we want to move to a more modern build system or stay for the foreseable future in the past (this statement should indicate my opinion on the subject :-)).
I maintain a branch with the meson integration that I keep rebasing on top of the v4l-utils master branch. You can find it at
git://linuxtv.org/pinchartl/v4l-utils.gitmesonI have also just posted the latest version of the integration patches in a v6 ([8]).
I have been using the meson integration for about 2 years now, and it has provided me with a much smoother experience than autoconf, both for native builds and cross builds. Long gone are the day where I had to fight autoconf and hack various Makefile.am to comment out pieces of the tree that would fail to compile properly and wouldn't want to get disabled through autoconf. These issues are most likely due to shortcomings in the autoconf usage in v4l-utils than problems with autoconf and automake themselves, but I quickly gave up on trying to fix that as meson just worked out of the box as intended.
This being said, I won't pretend that the current implementation would work perfectly for everybody. I twould thus like to get feedback on how to move forward.
Quoted 5 lines
1. Is there a general agreement that replacing autoconf is a good idea, provided that any technical issue in the proposed meson implementation (if any) can be fixed ? Or would it require fighting ophidiophobia and other non-technical issues that would make it a lost battle from the start ?
I did a quick check to see if it handles setting the date/build/sha correctly for some of the utilities I maintain (i.e. v4l2/cec-compliance needs to show the SHA of the commit it was built from), and that seems to be OK.
Given the fact that it is better at cross-compiling I have no objection to switching over.
It should be a complete switch, though. It's one or the other, not both.
If we do this, then I think we should try and prevent adding new libs or applications for a bit (one kernel cycle?) to make it easy to revert if we run into unexpected problems. And also bump the version number and ask Gregor to check that it builds fine for debian.
Regards,
Hans
Quoted 17 lines
2. What are the technical issues that still need to be solved (if any) to replace autoconf with meson ?
There's no need to wait for the media summit to start answering those questions, if we can resolve the issue before meeting face to face, we'll have more time to discuss other questions :-)
[1] https://lore.kernel.org/linux-media/[email protected]
[2] https://lore.kernel.org/linux-media/[email protected]
[3] https://lore.kernel.org/linux-media/[email protected]
[4] https://lore.kernel.org/linux-media/[email protected]
[5] https://lore.kernel.org/linux-media/[email protected]
[6] https://lore.kernel.org/linux-media/[email protected]
[7] https://lore.kernel.org/linux-media/[email protected]
[8] https://lore.kernel.org/linux-media/[email protected]
Hi Hans,
Quoted 49 lines · Hans Verkuil
Quoted 37 lines · Laurent Pinchart
Hello,
This mail (and hopefully mail thread) aims to prepare for the Media Summit 2022 discussion about finalizing the conversion of v4l-utils to meson.
The original port of v4l-utils to meson was done by Ariel D'Alessandro (big thanks for that !) and posted to the linux-media mailing list in April 2022 ([1]). Another RFC version followed ([2]), and the series then graduated to non-RFC ([3]) with new versions following ([4], [5] and [6]) until v5 ([7]) in May 2021. I believe it is time to decide if we want to move to a more modern build system or stay for the foreseable future in the past (this statement should indicate my opinion on the subject :-)).
I maintain a branch with the meson integration that I keep rebasing on top of the v4l-utils master branch. You can find it at
git://linuxtv.org/pinchartl/v4l-utils.gitmesonI have also just posted the latest version of the integration patches in a v6 ([8]).
I have been using the meson integration for about 2 years now, and it has provided me with a much smoother experience than autoconf, both for native builds and cross builds. Long gone are the day where I had to fight autoconf and hack various Makefile.am to comment out pieces of the tree that would fail to compile properly and wouldn't want to get disabled through autoconf. These issues are most likely due to shortcomings in the autoconf usage in v4l-utils than problems with autoconf and automake themselves, but I quickly gave up on trying to fix that as meson just worked out of the box as intended.
This being said, I won't pretend that the current implementation would work perfectly for everybody. I twould thus like to get feedback on how to move forward.
Quoted 5 lines
1. Is there a general agreement that replacing autoconf is a good idea, provided that any technical issue in the proposed meson implementation (if any) can be fixed ? Or would it require fighting ophidiophobia and other non-technical issues that would make it a lost battle from the start ?
I did a quick check to see if it handles setting the date/build/sha correctly for some of the utilities I maintain (i.e. v4l2/cec-compliance needs to show the SHA of the commit it was built from), and that seems to be OK.
Quoted 4 lines
Given the fact that it is better at cross-compiling I have no objection to switching over.
It should be a complete switch, though. It's one or the other, not both.
I agree, maintaining both would increase the maintenance burden and guarantee that bugs would creep in over time.
Quoted 4 lines
If we do this, then I think we should try and prevent adding new libs or applications for a bit (one kernel cycle?) to make it easy to revert if we run into unexpected problems. And also bump the version number and ask Gregor to check that it builds fine for debian.
I'm fine with that. Maintaining a meson branch on top of v4l-utils has been relatively low effort, there were occasional additions to v4l-utils, but they were easy to handle. I would think that reverting would be equally easy, if needed.
I will also be there to help addressing any bug in the build system.
Quoted 15 lines
Quoted 15 lines
2. What are the technical issues that still need to be solved (if any) to replace autoconf with meson ?
There's no need to wait for the media summit to start answering those questions, if we can resolve the issue before meeting face to face, we'll have more time to discuss other questions :-)
[1] https://lore.kernel.org/linux-media/[email protected]
[2] https://lore.kernel.org/linux-media/[email protected]
[3] https://lore.kernel.org/linux-media/[email protected]
[4] https://lore.kernel.org/linux-media/[email protected]
[5] https://lore.kernel.org/linux-media/[email protected]
[6] https://lore.kernel.org/linux-media/[email protected]
[7] https://lore.kernel.org/linux-media/[email protected]
[8] https://lore.kernel.org/linux-media/[email protected]
Signature · 3 lines
Regards, Laurent Pinchart
Quoted 63 lines · Laurent Pinchart
Quoted 49 lines · Hans Verkuil
Quoted 37 lines · Laurent Pinchart
Hello,
This mail (and hopefully mail thread) aims to prepare for the Media Summit 2022 discussion about finalizing the conversion of v4l-utils to meson.
The original port of v4l-utils to meson was done by Ariel D'Alessandro (big thanks for that !) and posted to the linux-media mailing list in April 2022 ([1]). Another RFC version followed ([2]), and the series then graduated to non-RFC ([3]) with new versions following ([4], [5] and [6]) until v5 ([7]) in May 2021. I believe it is time to decide if we want to move to a more modern build system or stay for the foreseable future in the past (this statement should indicate my opinion on the subject :-)).
I maintain a branch with the meson integration that I keep rebasing on top of the v4l-utils master branch. You can find it at
git://linuxtv.org/pinchartl/v4l-utils.gitmesonI have also just posted the latest version of the integration patches in a v6 ([8]).
I have been using the meson integration for about 2 years now, and it has provided me with a much smoother experience than autoconf, both for native builds and cross builds. Long gone are the day where I had to fight autoconf and hack various Makefile.am to comment out pieces of the tree that would fail to compile properly and wouldn't want to get disabled through autoconf. These issues are most likely due to shortcomings in the autoconf usage in v4l-utils than problems with autoconf and automake themselves, but I quickly gave up on trying to fix that as meson just worked out of the box as intended.
This being said, I won't pretend that the current implementation would work perfectly for everybody. I twould thus like to get feedback on how to move forward.
Quoted 5 lines
1. Is there a general agreement that replacing autoconf is a good idea, provided that any technical issue in the proposed meson implementation (if any) can be fixed ? Or would it require fighting ophidiophobia and other non-technical issues that would make it a lost battle from the start ?
I did a quick check to see if it handles setting the date/build/sha correctly for some of the utilities I maintain (i.e. v4l2/cec-compliance needs to show the SHA of the commit it was built from), and that seems to be OK.
Quoted 4 lines
Given the fact that it is better at cross-compiling I have no objection to switching over.
It should be a complete switch, though. It's one or the other, not both.
I agree, maintaining both would increase the maintenance burden and guarantee that bugs would creep in over time.
Quoted 4 lines
If we do this, then I think we should try and prevent adding new libs or applications for a bit (one kernel cycle?) to make it easy to revert if we run into unexpected problems. And also bump the version number and ask Gregor to check that it builds fine for debian.
Quoted 4 lines
I'm fine with that. Maintaining a meson branch on top of v4l-utils has been relatively low effort, there were occasional additions to v4l-utils, but they were easy to handle. I would think that reverting would be equally easy, if needed.
Also, if we already have a consensus among the audience of the media summit, we can remove this discussion point from the agenda and use the time for other topics.
Should I prepare a new version of the patches, removing autoconf/automake support on top ?
Quoted 17 lines
I will also be there to help addressing any bug in the build system.
Quoted 15 lines
Quoted 15 lines
2. What are the technical issues that still need to be solved (if any) to replace autoconf with meson ?
There's no need to wait for the media summit to start answering those questions, if we can resolve the issue before meeting face to face, we'll have more time to discuss other questions :-)
[1] https://lore.kernel.org/linux-media/[email protected]
[2] https://lore.kernel.org/linux-media/[email protected]
[3] https://lore.kernel.org/linux-media/[email protected]
[4] https://lore.kernel.org/linux-media/[email protected]
[5] https://lore.kernel.org/linux-media/[email protected]
[6] https://lore.kernel.org/linux-media/[email protected]
[7] https://lore.kernel.org/linux-media/[email protected]
[8] https://lore.kernel.org/linux-media/[email protected]
Signature · 3 lines
Regards, Laurent Pinchart
Quoted 6 lines · Laurent Pinchart
Also, if we already have a consensus among the audience of the media summit, we can remove this discussion point from the agenda and use the time for other topics.
Should I prepare a new version of the patches, removing autoconf/automake support on top ?
Yes, please. I'd like to see a version with the FIXUPs squashed in, and that makes it easy to revert if needed.
BTW, I think patch 1 can just be applied regardless, although v4l-utils.spec.in needs updating as well (README -> README.md).
Regards,
Hans
Quoted 10 lines · Hans Verkuil
Quoted 6 lines · Laurent Pinchart
Also, if we already have a consensus among the audience of the media summit, we can remove this discussion point from the agenda and use the time for other topics.
Should I prepare a new version of the patches, removing autoconf/automake support on top ?
Yes, please. I'd like to see a version with the FIXUPs squashed in, and that makes it easy to revert if needed.
https://lore.kernel.org/linux-media/[email protected]
Quoted 2 lines
BTW, I think patch 1 can just be applied regardless, although v4l-utils.spec.in needs updating as well (README -> README.md).
Done in v7.
Signature · 3 lines
Regards, Laurent Pinchart