www.digitalmars.com         C & C++   DMDScript  

D.gnu - small bugfix for std.socket; use reentrant gethostbyname

reply Downs <default_357-line yahoo.de> writes:
483,485c485,489
<         hostent* he = gethostbyname(toStringz(name));
<         if(!he)
<             return false;
---

         hostent he_buf;
         char[4096] buffer; int error;
         hostent *he;
         if (gethostbyname_r(toStringz(name), &he_dump, buffer.ptr,
buffer.length, &he, &error)!=0) return false;


Also, the old way would cause a problem .. because as the manpage for
gethostbyname will verify, if he != 0 doesn't mean there's no error ..
that's what h_errno is for. A return value of 0 from gethostbyname_r,
however, does guarantee there's no error.
(The problem with the return value would cause an error later on,
  where populate had to work with an invalid struct - that's how I
  found it in the first place)

Greetings --downs
Mar 18 2007
parent reply Downs <default_357-line yahoo.de> writes:
Downs schrieb:

 483,485c485,489
 <         hostent* he = gethostbyname(toStringz(name));
 <         if(!he)
 <             return false;
 ---
  >         hostent he_buf;
  >         char[4096] buffer; int error;
  >         hostent *he;
  >         if (gethostbyname_r(toStringz(name), &he_dump, buffer.ptr, 
 buffer.length, &he, &error)!=0) return false;
 
 Also, the old way would cause a problem .. because as the manpage for
 gethostbyname will verify, if he != 0 doesn't mean there's no error ..
 that's what h_errno is for. A return value of 0 from gethostbyname_r,
 however, does guarantee there's no error.
 (The problem with the return value would cause an error later on,
  where populate had to work with an invalid struct - that's how I
  found it in the first place)
 
 Greetings --downs


Hold on a sec, it looks like mingw has no reentrant gethostbyname_r ... argh x_x
Mar 18 2007
parent reply Downs <default_357-line yahoo.de> writes:
<snip some false information>

Looks like I have to apologize. From the manpage:


       The  gethostbyname()  and  gethostbyaddr() functions return the hostent
       structure or a NULL pointer if an error occurs.  On error, the  h_errno



But that is extremely odd, since I have a verified case
where gethostbyname would return a struct with an
_invalid_ h_aliases entry. Confusing.
Perhaps a threading issue? Should gethostbyname usage
be synchronized? I'll look into that.
                apologetic --downs
Mar 18 2007
parent Downs <default_357-line yahoo.de> writes:
Downs schrieb:

 <snip some false information>
 
 Looks like I have to apologize. From the manpage:
 
  >       The  gethostbyname()  and  gethostbyaddr() functions return the 
 hostent
  >       structure or a NULL pointer if an error occurs.  On error, the  
 h_errno
 
 
 But that is extremely odd, since I have a verified case
 where gethostbyname would return a struct with an
 _invalid_ h_aliases entry. Confusing.
 Perhaps a threading issue? Should gethostbyname usage
 be synchronized? I'll look into that.
                apologetic --downs


Yup, wrapping the gethostbyname function in a synchronized
block seems to fix this for me. Any reason not to do that?
  --downs
Mar 18 2007