Re: Optimization & factorization of ‘write’

[Ludovic Courtès]

Hi,

Mike Gran <address@hidden> writes:

>> >  But I tried it just now on HEAD, and it triggered a 'glibc detected
>> >  corrupteddouble-linked list' error.  Curious.
>> 
>> Hmm indeed.  Can  you come up with a reduced test case and perhaps a
>> backtrace or  something?
>> 
>
> in  scm_i_unistring_escapes_to_r6rs_escapes(), you can have a write
> off the end of a string when the buffer passed into the function contains
> only 4-digit hex unistring hex escapes, such as "\u1100".  The
> R6RS-escaped string will be longer "\x1100;" causing the memcpy at the
> end of the function to write off then end of the string.

Indeed, good catch!

I think commit f1ee6d54d219056c62d87a8e4a6b199162c946e8 hackily fixes
it.

The whole thing is hackish though: the conversion is inelegant, and it
assumes that BUF is in an ASCII-compatible encoding.  I think the right
way would be to have libunistring allow us to specify what we want
escapes to look like.

Thanks,
Ludo’.