1

[alpine-devel] what should be done with SA_INTERRUPT? (gpm 1.20.7)

Isaac Dunham
Details
Message ID
<20140716011646.GA1179@newbook>
Sender timestamp
1405473407
DKIM signature
missing
Download raw message
On alpine, gpm 1.20.7 doesn't build completely, so I've been poking at it.
There were numerous cases of sys/select.h being omitted, but the biggest
issue I've hit so far is the use of SA_INTERRUPT in a call to sigaction();
the purpose is to break out of select() without it getting restarted.
musl does not define or support SA_INTERRUPT, as far as I can tell (I
seem to recall that musl forces SA_RESTART).

This is in src/prog/gpm-root.y, which is the source for the "gpm-root"
utility.

Any idea what to do about this?
I could build gpm without it, but (a) I'd like to see the whole mess
build, and (b) I'm not sure how to tell make that it can skip gpm-root.

As a temporary fix, I'm using this patch:
-#if defined(__GLIBC__)
-   __sigemptyset(&childaction.sa_mask);
-#else /* __GLIBC__ */
-   childaction.sa_mask=0;
-#endif /* __GLIBC__ */
+   sigemptyset(&childaction.sa_mask);
+#ifdef SA_INTERRUPT
    childaction.sa_flags=SA_INTERRUPT; /* need to break the select() call */
+#endif


Thanks,
Isaac Dunham



---
Unsubscribe:  alpine-devel+unsubscribe@lists.alpinelinux.org
Help:         alpine-devel+help@lists.alpinelinux.org
---

[alpine-devel] Re: [musl] what should be done with SA_INTERRUPT? (gpm 1.20.7)

Rich Felker
Details
Message ID
<20140715190301.GG17402@brightrain.aerifal.cx>
In-Reply-To
<20140716011646.GA1179@newbook> (view parent)
Sender timestamp
1405450981
DKIM signature
missing
Download raw message
On Tue, Jul 15, 2014 at 06:16:47PM -0700, Isaac Dunham wrote:
> On alpine, gpm 1.20.7 doesn't build completely, so I've been poking at it.
> There were numerous cases of sys/select.h being omitted, but the biggest
> issue I've hit so far is the use of SA_INTERRUPT in a call to sigaction();
> the purpose is to break out of select() without it getting restarted.
> musl does not define or support SA_INTERRUPT, as far as I can tell (I
> seem to recall that musl forces SA_RESTART).

No, per POSIX, interrupting is the default behavior of sigaction;
SA_RESTART is needed to override this. The default you're thinking
about is for the signal() function which should not be used anyway.
There is no point at all in SA_INTERRUPT since it's required to be the
default for sigaction; I'm assuming it came from some pre-POSIX era
when the default was not clear.

> This is in src/prog/gpm-root.y, which is the source for the "gpm-root"
> utility.
> 
> Any idea what to do about this?
> I could build gpm without it, but (a) I'd like to see the whole mess
> build, and (b) I'm not sure how to tell make that it can skip gpm-root.
> 
> As a temporary fix, I'm using this patch:
> -#if defined(__GLIBC__)
> -   __sigemptyset(&childaction.sa_mask);
> -#else /* __GLIBC__ */
> -   childaction.sa_mask=0;
> -#endif /* __GLIBC__ */
> +   sigemptyset(&childaction.sa_mask);
> +#ifdef SA_INTERRUPT
>     childaction.sa_flags=SA_INTERRUPT; /* need to break the select() call */
> +#endif

Just remove the SA_INTERRUPT line entirely and (like you're doing) use
sigemptyset unconditionally.

Rich


---
Unsubscribe:  alpine-devel+unsubscribe@lists.alpinelinux.org
Help:         alpine-devel+help@lists.alpinelinux.org
---