Please add a repository_ctx method for reading files by Label

[jmillikin]

A repository_ctx can have Label attributes that point to files in other repositories (useful for tool plumbing), but such attributes are awkward to work with because there's no direct way to read their content.

The best workaround I've found is to use symlinks and run cat as a subprocess, which certainly deserves an entry in the Useless Use of Cat Award.

def _some_repo_rule(ctx):
  ctx.symlink(ctx.attr.genrule_cmd_file, "external-genrule-cmd.sh")
  cat_genrule_cmd = ctx.execute(["cat", "external-genrule-cmd.sh"])
  if cat_genrule_cmd.return_code != 0:
    fail("Failed to read genrule command %r: %s" % (
        ctx.attr.genrule_cmd_file, cat_genrule_cmd.stderr))
  genrule_content = cat_genrule_cmd.stdout

That whole mess could, ideally, be simplified down to this:

def _some_repo_rule(ctx):
  genrule_content = ctx.read_file(ctx.attr.genrule_cmd_file)

[ittaiz]

This sounds related to (though definitely not a duplicate of) #3438

[cgruber]

Has there been any progress on this, or prioritization? It's worse than the issue description when you consider windows compatibility, as "cat" isn't an external executable on that platform (it being a powershell alias to Get-Content, which is also not an external executable file. So for windows, one has to implement platform-specific toolchains to supply a call to powershell with that command, all to read text file contents with a repository_ctx.

[jmillikin]

PR: #7309

[dslomov]

What is the motivation for reading the file into Starlark rule?

[cgruber]

Getting metadata in that you can use to generate the repository more effectively. In my case, for example, getting teh appropriate maven metadata (pom files) to ensure that we generate dependencies properly in the generated bazel targets. Right now I'm using "cat" which doesn't work on windows. I don't want to have to set up a per-platform toolchain just to read a file.

[jmillikin]

I've been reading files in repository rules for two reasons:

1. Sometimes the choice of which artifacts to download is driven by a file from the main workspace. For example, Python or Node.JS projects save their download URLs to "lock files". The only way to access the content of a file from the main workspace is via the symlink hack described in the first post.

2. Some external dependencies require running a setup tool, which outputs files with non-deterministic paths. I want to inspect the log file generated by those tools to discover the paths, so they can be inserted into the generated BUILD file.

[jin]

+1, would be nice to have an OS-independent cat like feature in a repository rule.

[jin]

This came up in a discussion with @dkelmer earlier on again. Currently, to do cross-platform cat in a repository rule, we'll need to do this:

def _cat_file(repository_ctx, filepath):
    # For Windows, use cat from msys.
    cat = "C:\\msys64\\usr\\bin\\cat" if (_is_windows(repository_ctx)) else repository_ctx.which("cat")
    exec_result = repository_ctx.execute([cat, repository_ctx.path(filepath)])
    if (exec_result.return_code != 0):
        fail("Error while trying to read %s: %s" % (filepath, exec_result.stderr))
    return exec_result.stdout

It'll be nice to shorten this to content = repository_ctx.read(file).

[cgruber]

I mean, it's that or add toolchains. :)