[PATCH 1 of 2 STABLE RESEND] setup: do not select hg executable that prints unexpected warnings

Jun Wu quark at fb.com
Thu Jul 27 12:29:32 EDT 2017


Looks good to me. I also have a local fix that ignores warnings in both
places. It seems treating warnings as fatal might cause less surprises.

Excerpts from Yuya Nishihara's message of 2017-07-27 23:35:03 +0900:
> # HG changeset patch
> # User Yuya Nishihara <yuya at tcha.org>
> # Date 1500557557 -32400
> #      Thu Jul 20 22:32:37 2017 +0900
> # Branch stable
> # Node ID 26e41875da7be30b96b904ab3f52615cd19029b5
> # Parent  08f557c2b20e2ae98f57b76fcb2ea0d870cfdcba
> setup: do not select hg executable that prints unexpected warnings
> 
> Otherwise the subsequent hg.run() would fail. This factors out the filtering
> function so the same rule should apply.
> 
> diff --git a/setup.py b/setup.py
> --- a/setup.py
> +++ b/setup.py
> @@ -202,21 +202,25 @@ class hgcommand(object):
>      def run(self, args):
>          cmd = self.cmd + args
>          returncode, out, err = runcmd(cmd, self.env)
> -        # If root is executing setup.py, but the repository is owned by
> -        # another user (as in "sudo python setup.py install") we will get
> -        # trust warnings since the .hg/hgrc file is untrusted. That is
> -        # fine, we don't want to load it anyway.  Python may warn about
> -        # a missing __init__.py in mercurial/locale, we also ignore that.
> -        err = [e for e in err.splitlines()
> -               if not e.startswith(b'not trusting file') \
> -                  and not e.startswith(b'warning: Not importing') \
> -                  and not e.startswith(b'obsolete feature not enabled')]
> +        err = filterhgerr(err)
>          if err or returncode != 0:
>              printf("stderr from '%s':" % (' '.join(cmd)), file=sys.stderr)
> -            printf(b'\n'.join([b'  ' + e for e in err]), file=sys.stderr)
> +            printf(err, file=sys.stderr)
>              return ''
>          return out
>  
> +def filterhgerr(err):
> +    # If root is executing setup.py, but the repository is owned by
> +    # another user (as in "sudo python setup.py install") we will get
> +    # trust warnings since the .hg/hgrc file is untrusted. That is
> +    # fine, we don't want to load it anyway.  Python may warn about
> +    # a missing __init__.py in mercurial/locale, we also ignore that.
> +    err = [e for e in err.splitlines()
> +           if (not e.startswith(b'not trusting file')
> +               and not e.startswith(b'warning: Not importing')
> +               and not e.startswith(b'obsolete feature not enabled'))]
> +    return b'\n'.join(b'  ' + e for e in err)
> +
>  def findhg():
>      """Try to figure out how we should invoke hg for examining the local
>      repository contents.
> @@ -240,7 +244,7 @@ def findhg():
>          retcode, out, err = runcmd(hgcmd + check_cmd, hgenv)
>      except EnvironmentError:
>          retcode = -1
> -    if retcode == 0:
> +    if retcode == 0 and not filterhgerr(err):
>          return hgcommand(hgcmd, hgenv)
>  
>      # Fall back to trying the local hg installation.
> @@ -252,7 +256,7 @@ def findhg():
>          retcode, out, err = runcmd(hgcmd + check_cmd, hgenv)
>      except EnvironmentError:
>          retcode = -1
> -    if retcode == 0:
> +    if retcode == 0 and not filterhgerr(err):
>          return hgcommand(hgcmd, hgenv)
>  
>      raise SystemExit('Unable to find a working hg binary to extract the '


More information about the Mercurial-devel mailing list