On Sun, May 6, 2012 at 6:12 AM, David Bremner <david@tethera.net> wrote:
> I pushed this series, but it seems to have a test failure on the
> buildbot (it's fine on my Debian wheezey box)
>
> http://buildbot.notmuchmail.org/builders/master/builds/198/steps/tests/logs/problems
>
> Maybe you and Tom could try to figure out what is different about his
> environment?

The failure is on the new HTML reply test, and the difference in the
output is just indentation - the quoted HTML text has an extra three
spaces at the beginning of each line. So, it's almost certainly a
difference in emacs versions or packages. I'm running Ubuntu Oneiric
with emacs 23.3.

It looks like the relevant variable in mm is
mm-text-html-renderer-alist, which defines various methods of parsing
HTML. For me it's:

((w3 . mm-inline-text-html-render-with-w3)
 (w3m . mm-inline-text-html-render-with-w3m)
 (w3m-standalone . mm-inline-text-html-render-with-w3m-standalone)
 (links mm-inline-render-with-file mm-links-remove-leading-blank
"links" "-dump" file)
 (lynx mm-inline-render-with-stdin nil "lynx" "-dump" "-force_html"
"-stdin" "-nolist")
 (html2text mm-inline-render-with-function html2text))

I get the behavior seen on the buildbot if the only option available
on my machine is lynx, so I'm guessing that's what's happening.

I'm not sure what the best option is for fixing this. A few possible options:

* Only run the test if links, w3m, or w3m.el is available.
* Make the test accept differences in whitespace.
* Try to produce HTML for the test that renders the same in every method.

Opinions from anyone?

-- Adam