01-06-13 17.48, skrev Jarkko Hietaniemi p� [EMAIL PROTECTED] f�ljande:
> I have an idea on how could we approach the _r-mess piecemeal:
> introduce a new Configure use directive, for example -Dusereentrant,
> which would at first only add the use of the most common forms [1]
> of the _r functions (no Configure probes for the prototypes of the _r,
> even).
>
> [1] Or even for the first cut concentrate on *one* platform, like
> for example Linux, and get it first working there.
>
> I can add the Configure framework, but someone else will have
> to worry about things like marrying the state data of the _r
> functions and Perl's thread structures.
Excellent!
I think that sounds like a very good approach. I can take care of the non
configure parts for glibc.
I would suggest we start with gmtime/localtime, this is one of the functions
that is not safe, is used alot and also is probobly going to be used in
threaded enviroments. Somone will probobly nag me for this but the need for
the unix get* functions to be converted to _r is not as high since thier use
in multithreaded applications might not be so usefull. (I am mainly talking
about mod_perl 2.0 use here).
Should we still device a locking scheme for those functions which we haven't
made safe yet?
Artur
