Subject: Re: [boost] [conversion] Isolating the phantom file changes problem
From: Gavin Lambert (gavinl_at_[hidden])
Date: 2013-11-26 18:34:27
On 26/11/2013 20:18, Quoth Jürgen Hunold:
> On Tuesday, 26. November 2013 20:06:00 Gavin Lambert wrote:
>> On 26/11/2013 19:54, Quoth Jürgen Hunold:
>>> boost.svn$ svn proplist libs/pool/doc/images/mb1.svg
>>> Properties on 'libs/pool/doc/images/mb1.svg':
>>> boost.svn$ svn propget svn:mime-type libs/pool/doc/images/mb1.svg
>>> And image/* marks this as "binary". We had this discussion weeks ago with
>>> noch real decision.
>> Just curious: where is that rule coming from? If it's in the conversion
>> code (which seems likely), can it be changed to treat */*+xml as text
>> regardless of other rules?
> No, subversion will not do CRLF-conversion on binary files. This creates the
> same issues as explicit CRLF settings when doing the conversion on Unix. And
> the conversion script will have to use subversion commands in order to get the
> The reason for the "image/*" setting for svg was to have them displayed as
> images when viewed in a web browser. But those display settings can be better
> configured server-side. My experience is to have all text file eol-style
> "native" and mime-type "text/something" to get the best cross-platform
> integration. This is true even for .vcproj files as you can then script-edit
> them on Unix without problems.
I think you misunderstood my question. I understand why they were
labelled as image/svg+xml, and I understand why there's a "mime type
image/* in svn => binary attribute in git" rule.
I'm just suggesting that (if it makes sense with how the conversion tool
actually works -- I don't know much about it) a higher priority rule be
added to the conversion tool that says "mime type */*+xml in svn => text
attribute in git", which would then result in SVG files being marked as
text and all other files marked as images.
Though it'd still require the files to be dos2unix'd if the conversion
was run on Linux -- if it was run on Windows then the text attribute by
itself should fix the problem.
Another possibility might be to explicitly mark all text files in
.gitattributes with eol=lf and then run the conversion; I think that
would fix the repository blobs, and then the eol attribute can be
removed for actual use.