[Bug-wget] need some advice for 0 timeout code

[Tim Ruehsen]

Hi,

using gnutls with --timeout=0 fails. When trying to repair it, I stumbled over 
some code design flaws that I would like to fix.

[On 28.3.2012 we already had a short conversation to clarify that --timeout=0 
means "no timeout" = "INFINITE timeout".]

--timeout=0 sets timeout values like opt.read_timeout to 0, which in turn 
needs special care, since select_fd() takes a value of 0 as 'immediate return 
/ polling' and has no special value for "INFINITY".

There are three obvious ways to fix the issue once and for all:
a)
when given the timeout value 0, use INFINITY and special handle this value in 
select_fd() to call select with timeout NULL.
- INFINITY is defined in math.h which needs to be included in connect.c and 
init.c.
- cmd_time() maybe needs a clone to set a 0 value to INFINITY.
- checks like if (timeout) have to be changed into if (timeout!=INFINITY).

b)
when given the timeout value 0, use a very high timeout value like 100 years 
(maybe larger to handle future extrasolar communication ;-).
- since the code for timeout!=0 is very well tested since it is the normal 
case, we won't *need* any further changes.
- we could optionally get rid of the if (timeout) ... extra code in connect.c 
and gnutls (but get some extra calls to select()).

c) just fix gnutls.c.
- similar to a), but limited to gnutls.c and connect.c.
- gnutls seem to need NONBLOCKING sockets, so we would call select() / 
select_fd() anyways. Again unneccessary extra code to maintain...


I tend to like b) the most because it's elegent and improves maintainability.
Which solution would you prefer resp. which solution has a chance being 
accepted by the maintainer(s) ?

Tim