Uzbl

AFAICT gscanner is interpreting the '1' in "1b4" as the start of a number, and then returns a G_TOKEN_ERROR because it's surprised when 'b' appears.

A solution is to enable scan_string_sq, add this to expand_template():

else if(token == G_TOKEN_STRING) {
g_string_append(ret, (gchar *)g_scanner_cur_value(uzbl.scan).v_string);
}

and do set useragent = 'Mozilla/5.0 (X11; U; Linux x86_64; en-US; rv:1.9.1b4) Gecko/20090528 Gentoo Firefox/3.5b4'.

There doesn't seem to be a way to tell gscanner to ignore numbers. scan_string_dq might work too, but it interferes with the example status_format.

- the above patch is incomplete (it changes the user interface, this must be documented)
- Furthermore, you must now do things like:
set useragent = 'uzbl 7a (Webkit' WEBKIT_MAJOR.WEBKIT_MINOR.WEBKIT_MICRO)
which looks quite ugly. I hope we can figure out something cleaner that does not force the user to add quotes to variables that we already know are strings (see var_name_to_ptr and proto_var)

Agreed that that's ugly (although that wasn't a great example, the only thing that actually needed to be quoted was the 7a).

The commit linked below contains a replacement for gscanner. It's very basic, it doesn't do parsing of integers or floats or quoted strings or any of the other things that gscanner does. There's no UI change.

http://github.com/bct/uzbl/commit/202685865f8947c19baed2dfd539c145958c21ed

I would first like to be assured we cannot do this with the current approach before resorting to "home built" replacements.

Asked on gtk-list yesterday, still waiting for a response. I did find this thread, which says that it's not possible for floats:

http://markmail.org/message/eymgig6gsjxyf2id?q=gscanner&page=8

I think it's a fair assumption that the same applies to integers too (and a quick skim of gscanner.c appears to confirm it). It looks like gscanner wasn't really meant for a usecase as simple as ours.

Thanks bct.
Too bad no one responded on your thread (http://www.spinics.net/lists/gtk/msg15370.html)

But I think the good news is we can use rob's little language that also supports variables. So we can get rid of gscanner and don't need to write a replacement for it anymore :)

bct did the really, really nice job of moving all the expand_template() code to Rob's variable expansion stuff. This bug is fixed in my experimental, soon in master.