Regression with gdb.texinfo

[José Fonseca]

It seems that my first patch (replace execute_string by insert_string in
xml_insert_indexentry) caused a regression processing gdb.texinfo:

makeinfo  --docbook --ifinfo  -o gdb.xml gdb.texinfo
gdb.texinfo:6984: Unmatched }.
gdb.texinfo:11413: Unmatched }.
gdb.texinfo:11849: Unmatched }.
gdb.texinfo:11245: Unmatched }.
gdb.texinfo:11843: Unmatched }.
gdb.texinfo:11853: Unmatched }.
gdb.texinfo:11854: Unmatched }.
makeinfo: Removing output file `gdb.xml' due to errors; use --force to
preserve.

A breakpoint on pop_and_call_brace() in makeinfo.c shows:

Breakpoint 1, pop_and_call_brace () at makeinfo.c:2108
2108          line_error (_("Unmatched }"));
(gdb) bt
#0  pop_and_call_brace () at makeinfo.c:2108
#1  0x08056f46 in reader_loop () at makeinfo.c:2052
#2  0x08059dd5 in execute_string (format=0x0) at makeinfo.c:4016
#3  0x08061a20 in xml_insert_indexentry (entry=0x8070820 "type}", 
    node=0x8094308 "Type Checking") at xml.c:1407
#4  0x0804f675 in cm_printindex () at index.c:767
#5  0x08056c11 in read_command () at makeinfo.c:1905
#6  0x08056f0f in reader_loop () at makeinfo.c:2023
#7  0x08056377 in convert_from_loaded_file (name=0xbffffa9e "gdb.texinfo")
     at makeinfo.c:1573
#8  0x08054f15 in main (argc=6, argv=0xbffff8f4) at makeinfo.c:774
#9  0x40037e54 in __libc_start_main () from /lib/libc.so.6

The problem is that now trying to break the index term into a primary
and secondary term breaks in stuff like (gdb.texinfo:6984):

        @kindex set address@hidden, type}

since it will give "address@hidden" and "type}" as primary and secondary
respectively.

I don't what the best course of action here:
 - Try to detect cases like this one in xml_insert_indexentry. But of
   course that it's very difficult to handle every combination.
 - Add a xml_in_indexentry flag/counter, which would be incremented
   everytime a ", " is matched when executing the string, and
   opening/closing the respective PRIMARY, SECONDARY, etc. tag.
 - Simply do not try to break the index terms.
   
Revert to insert_string is not an option since it will break much more stuff.


BTW, with the latest changes in CVS the cvs.xml already is parsed
without any errors.

José Fonseca
__________________________________________________
Do You Yahoo!?
Everything you'll ever need on one web page
from News and Sport to Email and Music Charts
http://uk.my.yahoo.com