Re: POSIX bindings for GNAT

[emery@grebyn.com (David Emery)]

In article <4je5s7$ne3@tel.den.mmc.com>, hutchiso@epi.syr.ge.com wrote:

> However, there is one change which can cause problems for users. 
> Forest adds a new alternative to the type Signal_Masking, called Restart, 
> which specifies that the Posix.5 support should automatically handling 
> restarting of interupted actions. This is not a particularly bad extension
> and it is easy to support, unfortunately Forest makes this the default value
> for the signal masking parameter. This can provide suprises for client code 
> expecting the default value to be RTS_Signals or when moving to another
> implementation of Posix.5 which does not support the Restart option.

This was an option that we discussed when doing POSIX.5.  We concluded that
it was way too dangerous to support as a 'primitive'.  Consider a file
open that takes 1 minute, and you are on a system that has a 30 second timer.
THe open will never succeed (It'll always get EINTER).  With any sort of
auto-retry, the process will go into infinite loop.  ANd, this loop will be
very hard to understand, since the process will spend 99% of its time blocked
on the kernel.

            dave