[parted-devel] Parted on GNU Hurd based systems

[Debarshi 'Rishi' Ray]

> Have you considered keeping just the 3-argument sig-handlers
> and writing a one-argument stub handler for each that merely
> calls the 3-argument one (with a NULL 2nd arg).

Although it looked to be an elegant solution at first, I now realize
that it might create some complications on systems where 'sigaction'
is present but 'SA_SIGINFO' is not (like Hurd); or where 'sigaction'
itself is missing.

Due to the following block and a similar one in init_ui, the 'struct
sigaction' variables are only instantiated and assigned when
SA_SIGINFO is present in signal.h.
#ifdef SA_SIGINFO
static struct sigaction    sig_segv;
static struct sigaction    sig_int;
static struct sigaction    sig_fpe;
static struct sigaction    sig_ill;
#endif /* SA_SIGINFO */

Since the mere presence of 'sigaction' does not offset the absence of
'SA_SIGINFO' and we are stuck with one parameter handlers, the latter
is used as the defining entity. Moreover we do not have to separately
consider cases where 'sigaction' itself is absent. If there is no
'sigaction', then there is no 'SA_SIGINFO' as well.

When the one-argument stub calls the three-argument handler, we would
basically check whether the second parameter is NULL and act
accordingly. To avoid a compile time error on Hurd-like kernels we
would need to insert #ifdef...#else...#endif blocks in the
three-argument handler so that the mention of 'sigaction' and 'struct
sigaction' variables do not throw an error.

The argument that the #ifdef...#endif block be removed and 'struct
sigaction' variables be declared in all cases, so that insertion of
#ifdef...#else...#endif blocks are not necessary in the three-argument
handler is flawed because this will not take into consideration cases
where 'sigaction' itself is absent.

> That would avoid a lot of the duplication.

I feel this duplication is a price we can pay for a clearer code.
Having the control flow from a stub handler to the other; the checking
of the second parameter; the insertion of #ifdef...#else...#endif in
the three parameter handlers will reduce the readability.

In any case the duplication is very less, and having a separate
handler for the two cases keeps things a tad simpler.

Note: please find a 'git diff' attached. Though it compiles and works
on Linux kernel based systems, it is not complete. It will compile on
Hurd but not on 'sigaction' less systems. I thought it will make it
clear what I am trying to convey. Maybe I am taking a wrong aproach...

Happy hacking,
Debarshi
-- 
GPG key ID: 63D4A5A7
Key server: pgp.mit.edu
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ui.c.git-diff1
Type: application/octet-stream
Size: 9984 bytes
Desc: not available
Url : http://lists.alioth.debian.org/pipermail/parted-devel/attachments/20070306/d50aa2c4/ui.c.obj