Change TIdSocketListVCLPosix to use poll instead of select

[msccto]

fix for #445

I replaced TIdSocketListVCLPosix with the code you suggested in the #445

[msccto]

Hi,
is there any chance this PR is going to be merged?

thx
Matt

[zTorden]

nfds is defined as nfds_t, which is unsigned long int, and that's UInt64, not integer

[rlebeau]

Technically, nfds_t is simply defined as an "unsigned integer". Some platforms do use unsigned long, but there are others that use unsigned int instead.

[zTorden]

Just as i said t's defined in glibc as 'typedef unsigned long int nfds_t', but i do agree that there may be other definitions on different platforms, so i'd suggest using nativeuint here instead of integer

[rlebeau]

NativeUInt probably would not be appropriate either, because of the platforms that use unsigned int which doesn't change size between 32bit/64bit builds. There really needs to be separate platform-appropriate nfds_t definitions.

[zTorden]

you're right, and this is actually the case with glibc where it is always unsigned long int both on 32 and 64 bit. still maybe unsigned long int itself differs on 32bit, can't really tell. but on ubuntu 22.04 which is always 64bit it's UInt64, i've checked that. but who knows, maybe it's only valid for gcc and not some other compiler, or is dependent on some 'compatibility' compiler options.

[zTorden]

as for the de-facto standard System V AMD64 calling convention implemented on most (if not all) *nixes it's safe to be defined as integer (at least when there are less then 2^32 fds allocated), because even when it is passed via EDI register, the higher bits of RDI are being zeroed rendering it a valid 64bit value. e.g. after these operations

mov rdi, $ffffffffffffffff
mov edi, 0

the value of rdi would be 0 and not $ffffffff00000000

[zTorden]

but this is only valid for register-passed parameters (i.e. first 6 function parameters), if it's passed via stack (7+ parameter), or as a part of structure, or referenced by pointer it will cause trouble

[rlebeau]

In general, yes. But that should not be a problem in THIS case since poll() has only 3 parameters and they are all simply types (1 pointer and 2 integers). The 1st parameter is a pointer to a struct, but that does not affect the nfds_t parameter in question.