 |
lua-l archive |
|
|
||
|
Mark Edgar wrote:
I suppose I could be more specific and more helpful:
liolib.c needs to call pclose() instead of fclose() when the
file:close() metamethod is called. This is handled by aux_close:
http://www.lua.org/source/5.1/liolib.c.html#aux_close
The __close field is stored in the environment of each popen'ed
userdatum. This is done automatically because popen'ed userdata are
created only in io_popen() whose environment table is set explicitly
in luaopen_io().
Similarly, you could set an environment table on your function which
creates pipe userdata. Your get/set_timeout metamethods could use the
presense of this environment table (or a certain field contained in
it) to determine if a given userdata is indeed a pipe or something
else.
Note that if you want APR pipes to have the same semantics as
io.popen'ed pipes, then most of this work is already done for you ...
-Mark
Thanks for your detailed explanation. Note that I already have liolib.c opened next to apr/src/io_file.c permanently because I'm making the interface compatible with Lua's io library and liolib.c is a valuable source of inspiration.
You don't think it's bad style to add (get|set)_timeout to the file userdata methods? Not trying to criticize, just interested in your opinion.
For now I've decided to use a different metatable + __index for pipes, which means the (get|set)_timeout methods are only available on pipe userdata. This also means all argument validation in the file/pipe io library needs the luaL_checkudata_ex() helper I posted earlier in this thread. I'm not yet sure whether I prefer this or your suggestion above.
Thanks again to everyone for their input! - Peter Odding