Exclude a package when building all from sources

[uilianries]

We have 3 different ways to execute it, to illustrate:

Build all from sources, except zlib:

conan install boost/1.75.0@ --build=!zlib --build=*
conan install boost/1.75.0@ --build=!zlib --build

Do not build anything from sources. --build= means '' package, which is invalid.

conan install boost/1.75.0@ --build=!zlib --build=

Changelog: Feature: Skip package when building all package from sources at once using --build=!<package> syntax.
Docs: conan-io/docs#2023
closes #8442

• Refer to the issue that supports this Pull Request.
• If the issue has missing info, explain the purpose/use case/pain/need that covers this Pull Request.
• I've read the Contributing guide.
• I've followed the PEP8 style guides for Python code.
• I've opened another PR in the Conan docs repo to the develop branch, documenting this one.

^{Note: By default this PR will skip the slower tests and will use a limited set of python versions. Check here how to increase the testing level by writing some tags in the current PR body text.}

[uilianries]

If --build, --build= or --build=* is passed, build ALL from sources.

[uilianries]

--build= should be consider equal to --build. Otherwise, --build --build=foo will be parsed as --build=foo only.

[memsharded]

While I understand the current behavior is not consistent, and it would be better to fix it, I think that this will be an issue, breaking users that for some reason rely on the current behavior, and unveiling other issues. Needing to do patch releases one after the other, etc. The same as when we changed the ignore_case=True in the FileCopier, which after several fixes still have some open issues. And it was "right" to fix it, but not worth the effort.

I would say that this case is similar. It is better to wait until Conan 2.0 to resolve these ambiguities (I haven't seen users reporting this), and if we want to implement now the feature request that the title is suggesting: Exclude a package when building all from sources, that might still be possible introducing the --build=!pattern, or --build-exclude=pattern or something like that.

[uilianries]

Indeed. Let's keep it for 2.0.

While --build=!pattern is more intuitive, our current argparse usage does not allow it. If we pass --build --build=anything, the argparse returns ['anything'], it discards --build because is None. However, if we pass --build= --build=anything, argparse returns ['anything', ''], which fits for --build=!anything.

So to avoid this kind of confusion, I would prefer a new argument, as you mentioned --build-exclude.

UPDATE:

Thinking again, we can accept --build=!foo too, it doesn't matter. If we pass --build --build=!foo or --build --build-exclude=foo, argparse will ignore --build.

[uilianries]

Without it, we won't be able to know if --build has been passed together with another --build=foo or --build=!foo. Argparse excludes None entries. As '*' is considered build all from sources and --build has the same usage, I don't think it will break Conan contract.

[memsharded]

Having a exclude pattern shouldn't be tied to building from source everything. You could have --build=mypkgs* --build=!mypkgs_special and it should build all packages starting with mypkgs but not the excluded one.

Please try to avoid as much as possible to particularize a generic thing as Extender with custom logic from one of its consumer, it is an anti-pattern and should be used only in case of high necessity.

[memsharded]

I also think this might still be changing behavior, and this will probably bring issues, better to avoid it.

[uilianries]

By default, when we pass --build --build=!foo, our parse only returns !foo, so doesn't matters whether --build is listed as command argument or not. This new conditional, only accepts --build=! when --build is present too. Or, we can create an implicit rule, like if you passed only --build=!foo, it means build all from sources, except for foo.

[memsharded]

I would prefer it to be explicit, and move in the direction of Conan 2.0 CLI than adding it magically. So we could document that --build=!xxx requires other patterns to be positively defined, even if --build=*. Better than customize the Extender class with specific attribute logic (in other command --build could mean a different thing (actually it does in conan build --build command, and this is a risk of failure.

[uilianries]

-build=!xxx requires other patterns to be positively defined

by contrary, we don't need to define an extra argument, without this workaround. If you agree, I'll remove it and keep only --build=!xxx, which is enough for us.

[memsharded]

Once the exclude_patterns has been computed, we don't need the ! anymore. It seems it would be better to clean it directly while computing self._excluded_patterns

[uilianries]

My idea was printing to user that a package could be wrong named. If we don't keep that list, we can't provide such information.

[memsharded]

What do you mean? I don't understand. We can keep that list, but in any case the ! is syntactic sugar, lets call it --build-exclude and that will remove all ambiguities.

[uilianries]

Oh I see! Indeed if we use a new argument instead of !, we don't need to parse the value. Okay, if we will accept --build-exclude, I prefer accepting only it and excluding ! to avoid any ambiguous behavior.

[memsharded]

Shouldn't this use self._excluded_patterns? Seems a bit confusing.

[uilianries]

It's a possibility indeed, but the idea here is comparing if a pattern matches with excluded pattern, if so, exclude it from exclude patterns list and return False to say: Don't build this package from sources. All remaining patterns in excluded list will be printed to say: These packages don't match to a pattern to excluded.

[uilianries]

@memsharded I'm not going to say that we traded 6 for half a dozen, but add a new argument for --build-exclude and forwarding it, inflicted more files than before. Well, at least we didn't touch arg parser.

We still need exclude_pattern list, because only when checking "force", the package reference is available for validation. Take a look and tell me if you see a point which can be simplified. Honestly, I don't like to forward and forward a parameter just for a simple feature.

[memsharded]

@memsharded I'm not going to say that we traded 6 for half a dozen, but add a new argument for --build-exclude and forwarding it, inflicted more files than before. Well, at least we didn't touch arg parser.

We still need exclude_pattern list, because only when checking "force", the package reference is available for validation. Take a look and tell me if you see a point which can be simplified. Honestly, I don't like to forward and forward a parameter just for a simple feature.

Yes, I don't love this either... too much for such a feature. Need to re-think this one.

[uilianries]

@memsharded Good to go