external_diff should not assume diff is gnu diff

On Tue, 2009-01-06 at 02:52 +0000, Martin Pool wrote:
> Public bug reported:
>
> affects bzr
> tags diff bsd
> status confirmed
> importance medium
>
> bzr diff when running an external diff assumes it understands the gnu
> option --binary. bsd diff does not.

I'm not sure this is a bug; its reasonable to depend on certain
capabilities in external tools - bzr doesn't claim to need 'posix diff'.

-Rob
--
GPG key available at: <http://www.robertcollins.net/keys.txt>.

  summary external_diff should not assume diff is gnu diff

On 6 Jan 2009, Robert Collins <email address hidden> wrote:

> I'm not sure this is a bug; its reasonable to depend on certain
> capabilities in external tools - bzr doesn't claim to need 'posix diff'.

Having this feature just break is buggy. Various fixes are possible eg:

 * document that 'diff' must be GNU diff

 * run gdiff or gnudiff or whatever it's called

 * only use --binary on platforms where we reasonably expect that diff
   will support it

--
Martin

  summary "external_diff should not assume diff is gnu diff"

On 6 Jan 2009, Robert Collins <email address hidden> wrote:

> I'm not sure this is a bug; its reasonable to depend on certain
> capabilities in external tools - bzr doesn't claim to need 'posix diff'.

Having this feature just break is buggy. Various fixes are possible eg:

 * document that 'diff' must be GNU diff

 * run gdiff or gnudiff or whatever it's called

 * only use --binary on platforms where we reasonably expect that diff
   will support it

--
Martin

On Tue, 2009-01-06 at 09:13 +0000, Martin Pool wrote:
> Having this feature just break is buggy. Various fixes are possible eg:
>
> * document that 'diff' must be GNU diff
>
> * run gdiff or gnudiff or whatever it's called
>
> * only use --binary on platforms where we reasonably expect that diff
> will support it

I'd be inclined to just document that we need a diff supporting
--binary.

-Rob

--
GPG key available at: <http://www.robertcollins.net/keys.txt>.

That said, we only really need it on systems that do weird things with line endings, like Mac and Win32. I don't think --binary does anything for Linux or BSD.

So we could only supply the flag if "sys.platform in ('win32', 'cygwin')" and possibly 'darwin'. Though I'm pretty sure on darwin line-endings are not munged either.