Using GCC 3.3.1 on QNX NC 6.3.0

[Evan Hillas]

Colin Burgess wrote:

Danger - we don't support SA_RESTART so that was why it wasn't defined.
So you may have problems with interrupted kernel calls.

What constitutes an interrupted kernel call? I can guess at something like a wait being aborted. And what is likely to happen without this support?


Will Robertson

[Colin Burgess]

If a signal is dropped on you then the kernel call (say MsgReceivev)
will return with -1 and errno==EINTR

SA_RESTART is a unixism to tell the kernel to restart the kernel call as
if nothing happened.

Evan Hillas wrote:

Colin Burgess wrote:

Danger - we don't support SA_RESTART so that was why it wasn't
defined. So you may have problems with interrupted kernel calls.


What constitutes an interrupted kernel call? I can guess at something
like a wait being aborted. And what is likely to happen without this
support?


Will Robertson

--
cburgess@qnx.com

[John Nagle]

Evan Hillas wrote:

Evan Hillas wrote:

parse.C: In function `bool parse_int(const char*, const char*, int&)':
parse.C:68: error: invalid conversion from `const char*' to `char*'

Lines 67 and 68 from parse.C ...

bool parse_int(const char* buf, const char* tag, int& x) {
char* p = strstr(buf, tag);


Oh, click!, it is returning a pointer to a const char which means p
should be declared as such. Hmm, I just assumed they would have
something like that all nice and clean in such, supposedly, portable code.

Yes, you should declare p as "const char* p".

But unless you're compiling in C++ mode, the definition of "strstr"
should be

char *strstr(const char *s1, const char *s2);

This is a constness violation, but standardized as such in IEEE Std 1003.1-2001
for backwards compatibility with code that predates the
addition of "const" to C.

In C++ mode, there are two forms of "strstr", const and non-const.
Only C++ supports such overloading. So in C++, if you
put in "const" arguments to strstr, you get a const output,
which looks like what you're seeing.

So if you're getting that error, I suspect you've somehow
enabled C++ mode in "gcc". Which isn't necessarily bad;
good C code should compile in C++ mode.

Or maybe the Gnu people decided to pull that old compatibility hack
out.

John Nagle

[David Bacon]

This "unixism" (SA_RESTART) is also an XSI extension.

Are we ever planning to support it?

Meanwhile, how can one distinguish between a system which doesn't
support it at all (like Neutrino) and one which just doesn't declare
SA_RESTART because automatic resumption is already the semantics
(like old BSD-based, non-XSI systems)?

dB

- - - - - - - - - - - - - - - - - - - - - - - - - - - -
Colin Burgess wrote ~ Thu, 16 Jun 2005 20:13:19 -0400:

If a signal is dropped on you then the kernel call (say MsgReceivev)
will return with -1 and errno==EINTR

SA_RESTART is a unixism to tell the kernel to restart the kernel call as
if nothing happened.

Evan Hillas wrote:
Colin Burgess wrote:

Danger - we don't support SA_RESTART so that was why it wasn't
defined. So you may have problems with interrupted kernel calls.


What constitutes an interrupted kernel call? I can guess at something
like a wait being aborted. And what is likely to happen without this
support?


Will Robertson

--
cburgess@qnx.com

[Colin Burgess]

David Bacon wrote:

This "unixism" (SA_RESTART) is also an XSI extension.

Are we ever planning to support it?

I don't believe we have any plans.

Meanwhile, how can one distinguish between a system which doesn't
support it at all (like Neutrino) and one which just doesn't declare
SA_RESTART because automatic resumption is already the semantics
(like old BSD-based, non-XSI systems)?

I'm not sure. Some standards guy might be ablee to comment. :v)

dB

- - - - - - - - - - - - - - - - - - - - - - - - - - - -
Colin Burgess wrote ~ Thu, 16 Jun 2005 20:13:19 -0400:

If a signal is dropped on you then the kernel call (say MsgReceivev)
will return with -1 and errno==EINTR

SA_RESTART is a unixism to tell the kernel to restart the kernel call as
if nothing happened.

Evan Hillas wrote:

Colin Burgess wrote:


Danger - we don't support SA_RESTART so that was why it wasn't
defined. So you may have problems with interrupted kernel calls.


What constitutes an interrupted kernel call? I can guess at something
like a wait being aborted. And what is likely to happen without this
support?


Will Robertson

--
cburgess@qnx.com

--
cburgess@qnx.com