build: Support 'rename' in configure_file, build_target and custom_target

[keith-packard]

This allows built results from any build target, custom target or configure file to be renamed during installation, just as with install_data.

This feature is motivated by the need to install multiple files from a single source directory into different install directories but with the same basename. In my case, I build dozens of 'libc.a' files which need to be installed into the toolchain's multilib heirarchy correctly.

This replaces #11954 and I can use this instead of #4037.

[keith-packard]

In getting the tests working, I discovered that shared libraries can generate additional files, like the Windows import library, which aren't listed in the outputs array. As you can see from this test.json file, those get installed using the original names, without any rename support.

I could:

1. Ignore this
2. Have the rename list include entries for these additional files
3. Make the rename value affect only the directory and basename, then generate the actual names using the directory, prefix, basename and suffix much like the current code does for the target name to filename mapping.

Other than this issue, I think this MR is ready to go; it's passing all of the tests (save for a couple which seem to just have CI issues?).

[keith-packard]

I missed a case that I need -- configure_file. I've added another commit which fixes that.

[dcbaker]

I believe that the additional windows files the .lib and .pdb must have the same name as the .dll to be automatically loaded, I think that Windows .exe files can also generate a .pdb, and those need to match the name of the .exe.

[keith-packard]

I believe that the additional windows files the .lib and .pdb must have the same name as the .dll to be automatically loaded, I think that Windows .exe files can also generate a .pdb, and those need to match the name of the .exe.

If so, then just taking the rename value and editing that to the correct form might be a reasonable first approximation. Would it be OK if the first version of this feature simply disallowed use of rename when generating windows shared libraries?

[dcbaker]

Hmmm... IIUC, then we would just need to install the .lib and .pdb with the same name but .pdb, though maybe we have someone with more Windows expertise?

skipping Windows might be a possibility if getting it working is too hard, but I'd like to at least try to get it working.

It would also need to exclude Vala because there are implicit outputs from that too.

[keith-packard]

Hmmm... IIUC, then we would just need to install the .lib and .pdb with the same name but .pdb, though maybe we have someone with more Windows expertise?

"it's complicated". Cygwin uses a 'cyg' prefix rather than 'lib', while still using '.dll.a' for various reasons.

skipping Windows might be a possibility if getting it working is too hard, but I'd like to at least try to get it working.

Instead of attempting to automate this renaming, I've just added two new keyword args, 'import_rename' and 'debug_rename' that allow the user to explicitly specify the install names for those two files. This ensures that the user has complete control over all of the installed filenames.

It would also need to exclude Vala because there are implicit outputs from that too.

Vala uses the existing 'rename' mechanism as it extends the 'output' array with the new files. That means the user will already have to make the 'rename' array contain entries for the additional files.

A couple of options I considered:

• Implicit names for the import library and debuginfo files based upon the provided rename value. This would have been a bit tricky to make sure the 'correct' part of the rename value was used.
• Switching the rename values to just the basename and applying the name_prefix and name_suffix values to the main install file as well as applying the same editing rules to construct suitable import library and debuginfo file names. This would have made these new rename uses different from existing rename uses in the system, and that seemed like a user trap.
• Adding these new rename values to the existing rename list. That would have placed an implicit ordering on the list, which contrasts with mesons use of keyword arguments when order is not relevant. New keyword arguments make it easy to tell what the script is going to generate without having to know the order of the rename list.

[bonzini]

Can you rebase, hoping that will improve CI results a bit?

[keith-packard]

I've rebased both this and #14002. As I mentioned over there, I greatly prefer #14002 to this kludge, but this would at least let me get my multi-target clang builds working.