win32text on non-windows platforms

Mark Hammond skippy.hammond at
Fri Aug 21 21:03:49 CDT 2009

Hi all,
   I'm working with Dirkjan on some win32text enhancements suitable for 
Python to use when it migrates to hg, and as part of taking some first 
baby steps I've struck an issue I need some advice on.

My understanding is that the win32text extension is, in theory, suitable 
for use on non-Windows platforms.  However, the existing test-win32text 
assumes that even on non-windows platforms, files marked to be processed 
by win32text have Windows line endings.

For example, consider this portion of test-win32text.out:

| 3 files updated, 0 files merged, 0 files removed, 0 files unresolved
| hello<NUL><CR><LF>
| some<CR><LF>text<CR><LF>
| rem empty<CR><LF>

In the first line of data, the file is a treated as 'binary' so is 
correct everywhere.  However, the last 2 files are testing the EOL 
semantics for text files, and as a result, the output shown above is 
correct for Windows, but is *not* correct for Linux - Linux should see 
native line endings for files processed as text, especially if storing 
these rules in a versioned file becomes reality.

Ideally, we should be able to test the 'native' semantics, as well as 
the 'other' semantics regardless of what OS we are on - but at the 
moment we seem to be explicitly testing the wrong thing everywhere *but* 
Windows.  Attempting to run these tests on Windows fails in lots of 
other dramatic ways such that the fact it *should* pass on Windows is 
small consolation for reality :)  It isn't quite clear where I should 
start here...

All advice welcome!



More information about the Mercurial-devel mailing list