Do not change the locale globally, else things will break in weird and
wonderful ways.
Introduce utils/locale.[ch], which provide locale-specific wrappers for various
functions (currently just the <ctype.h> ones).
Fix up the few places I can see that actually require that the underlying
locale is paid attention to.
Some notes:
1) The GTK frontend code has not been touched. It is possible that reading of
numeric values (e.g. from the preferences dialogue) may break with this
change, particularly in locales that use something other than '.' as their
decimal separator.
2) The search code is left unchanged (i.e. assuming a locale of "C").
This may break case insensitive matching of non-ASCII characters.
I doubt that ever actually worked, anyway. In future, it should use
Unicode case conversion to achieve the same effect.
3) The text input handling in the core makes use of isspace() to detect
word boundaries. This is fine for western languages (even in the C locale,
which it's currently assuming). It will, however, break for CJK et. al.
(this has always been the case, rather than being a new issue)
4) text-transform uses locale-specific variants of to{lower,upper}. In future
this should probably be performing Unicode case conversion. This is the
only part of the core code that makes use of locale information.
In future, if you require locale-specific behaviour, do the following:
setlocale(LC_<whatever>, "");
<your operation(s) here>
setlocale(LC_<whatever>, "C");
The first setlocale will change the current locale to the native environment.
The second setlocale will reset the current locale to "C".
Any value other than "" or "C" is probably a bug, unless there's a really
good reason for it.
In the long term, it is expected that all locale-dependent code will reside in
platform frontends -- the core being wholly locale agnostic (though assuming
"C" for things like decimal separators).
svn path=/trunk/netsurf/; revision=4153
Aside from a number of instances of const being cast away (mostly relating to the urldb, which is correct to only export const data) this now builds warning-free with GCC 4 on x86, which is nice.
svn path=/trunk/netsurf/; revision=3868
* Only apply presentational HTML attributes if no more
important CSS has been set for the property. (NetSurf used
to be a bit hit-and-miss when presentational markup and
CSS were mixed.)
* Change table cellpadding and border handling to happen as
soon the boxes styles are available, rather than after the
whole table has been constructed. Also fix default table
border colour.
* Improve handling of CENTER tag and ALIGN attribute. These
could not be correctly supported in the default CSS file,
so block level element alignment is now done during box
construction. (Fixes#1891379, #1824492, #1723853)
Form improvements:
* Small MAXLENGTH values on text inputs now reduce element
width. (Fixes#1894854)
* Prevent select option text from wrapping.
svn path=/trunk/netsurf/; revision=3866
NetSurf includes are now done with ""s and other system includes with <>s as C intended.
The scandeps tool has been updated to only look for ""ed includes, and to verify that the
files exist in the tree before adding them to the dependency lines. The depend rule has
therefore been augmented to make sure the autogenerated files are built before it is run.
This is untested under self-hosted RISC OS builds. All else tested and works.
svn path=/trunk/netsurf/; revision=3307
Actually process form inputs which have been styled display:none;
This needs revisiting after 1.0 as the following will still break:
<form ..>
<div style="display:none;">
<input type="hidden" name="foo" value="bar"/>
</div>
<input type="submit" name="submit" value="submit"/>
</form>
The children of the div are not processed (which is correct for display
purposes, but results in the hidden input being ignored entirely). A
more correct fix would be to perform form input -> gadget creation
orthogonally from box tree generation; then styling will have no effect.
svn path=/trunk/netsurf/; revision=3236
John Mark Bell