Hi Dennnis/Rohit,

Thanks for your responsesCould you please explain me how does HPUX
pthread_cancel works in case of blocking system calls. EgIf i am in
middle of read system call and get a cancellation signal. read will return
with EINTR am i right ?and after that cancel signal still exists
or is said to be deliveredcoz If i do a testcancel after that I can
still see that signal

thanks in advance.

9/13/06, Manish Katiyar <mkatiyar (AT) gmail (DOT) comwrote:

Hi Dennis,

Thanks for the responseyes main thread should not be cancelled but I
see this behaviour when i create a thread alsojust to keep the code
short I cancelled the main thread.

"I assume that since you aren't calling any libc functions, there is no
place where synchronous cancels can be caught."

I partially agree with this coz as I know libc functions just act as
additional cancellation points. Even if they are not there and thread
behaviour is deferred, signal should be delivered to the thread at some
point and this is what I expect.

Any other suggestions ?
>
>
>

Hi Rohit,

Below is my code snippet and its output

#include<stdio.h>
#include<signal.h>
#include<pthread.h>

pthread_t stid,ftid;

void hndlr(void *a){
int oldstate;
printf("I am in pthread handler\n");
return ;
}
void *fstart(void *tid){
printf("Sending cancel signal to another thread \n");
pthread_cancel(stid);
return NULL;
}

void *sstart(void *tid){
char buff[100];
int n;
pthread_cleanup_push(hndlr,NULL);
printf("\nI have pushed a function\n");
printf("Going to block while read from stdin\n");
n=read(0,buff,100);
#ifdef A
pthread_testcancel();
#endif
pthread_cleanup_pop(0);
return NULL ;
}

int main(){
pthread_attr_t attr;
pthread_attr_init(&attr);
pthread_create(&stid,&attr,sstart,NULL);
sleep(3);
pthread_create(&ftid,&attr,fstart,NULL);
pthread_join(stid,NULL);
return 0;
}

/tmp/katiyarcc -o HP -lpthread HP.c
/tmp/katiyar./HP

I have pushed a function
Going to block while read from stdin
Sending cancel signal to another thread
/tmp/katiyarcc -DA -o HP -lpthread HP.c
/tmp/katiyar./HP

I have pushed a function
Going to block while read from stdin
Sending cancel signal to another thread
I am in pthread handler

Thanks in advance

9/15/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish:

From what you've written, if you're doing a blocking read(2) call, and a
cancellation is posted, it would deliver it and like you said, read(2)
will return and reflect EINTR in the errno. BUT, if its just executing
read(2) call, but not blocked, it would depend on whether cancellation
is delivered to it or not, since it would deliver to it only in certain
states.

cancellation is delivered, testcancel will not catch it. Could you
share the code snippet that you're trying, with us?

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish Katiyar
Sent: Friday, September 15, 2006 11:45 AM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Dennnis/Rohit,

Thanks for your responsesCould you please explain me how does HPUX
pthread_cancel works in case of blocking system calls. EgIf i am in
middle of read system call and get a cancellation signal. read will
return with EINTR am i right ?and after that cancel signal
still exists or is said to be deliveredcoz If i do a testcancel
after that I can still see that signal
--
thanks in advance.
>
>
>
9/13/06, Manish Katiyar <mkatiyar (AT) gmail (DOT) comwrote:

Hi Dennis,

Thanks for the responseyes main thread should not be cancelled
but I see this behaviour when i create a thread alsojust to
keep the code short I cancelled the main thread.

"I assume that since you aren't calling any libc functions, there is
no place where synchronous cancels can be caught."

I partially agree with this coz as I know libc functions just act as
additional cancellation points. Even if they are not there and thread
behaviour is deferred, signal should be delivered to the thread at
some point and this is what I expect.

Any other suggestions ?
>
>
>

Hi Rohit,

I am testing it on both 11.11 PA and 11.23 PAand on both it doesn't
work without compiling it with -DAIs there any patch issueI have the
patch 37418 applied on my 11.23 machine.

/home/katiyar./a.out

I have pushed a function
Going to block while read from stdin
Sending cancel signal to another thread
/home/katiyaruname -a
HP-UX hpsarpc1 B.11.11 U 9000/800 1579700508 unlimited-user license
/home/katiyar>

/tmp/katiyarcc -o HP -lpthread HP.c
/tmp/katiyar./HP

I have pushed a function
Going to block while read from stdin
Sending cancel signal to another thread
/tmp/katiyaruname -a
HP-UX holmium B.11.23 U 9000/800 1759078586 unlimited-user license

Thanks in advance

9/15/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish: I've tried the program on both 11.11(PA) and 11.23 (IA) servers.
Both the cases, the cleanup handler is called, with and without -DA
options.

Also, a small correction to what you'd stated in your mail previously
and I'd asserted that to be true - when you post the cancellation, the
thread doesn't return from the cancellation point, the thread is
terminated thereafter. If there is a registered cleanup handler, the
termination would be post the action of the handler.

The output I got on my machines is pasted below:

$ uname -a
HP-UX rocket B.11.11 U 9000/800 187494646 unlimited-user license
$ cc -o a -lpthread a.c
$ ./a

I have pushed a function
Going to block while read from stdin
Sending cancel signal to another thread
I am in pthread handler
$ cc -DA -o a -lpthread a.c
$ ./a

I have pushed a function
Going to block while read from stdin
Sending cancel signal to another thread
I am in pthread handler

Are you on 11.11(PA), 11.23 (IA) or 11.0?

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish Katiyar
Sent: Friday, September 15, 2006 3:19 PM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Rohit,

Below is my code snippet and its output
--

#include<stdio.h>
#include<signal.h>
#include<pthread.h>

pthread_t stid,ftid;

void hndlr(void *a){
int oldstate;
printf("I am in pthread handler\n");
return ;
}
void *fstart(void *tid){
printf("Sending cancel signal to another thread \n");
pthread_cancel(stid);
return NULL;
}

void *sstart(void *tid){
char buff[100];
int n;
pthread_cleanup_push(hndlr,NULL);
printf("\nI have pushed a function\n");
printf("Going to block while read from stdin\n");
n=read(0,buff,100);
#ifdef A
pthread_testcancel();
#endif
pthread_cleanup_pop(0);
return NULL ;
}

int main(){
pthread_attr_t attr;
pthread_attr_init(&attr);
pthread_create(&stid,&attr,sstart,NULL);
sleep(3);
pthread_create(&ftid,&attr,fstart,NULL);
pthread_join(stid,NULL);
return 0;
}

/tmp/katiyarcc -o HP -lpthread HP.c
/tmp/katiyar./HP

I have pushed a function
Going to block while read from stdin
Sending cancel signal to another thread
/tmp/katiyarcc -DA -o HP -lpthread HP.c /tmp/katiyar./HP

I have pushed a function
Going to block while read from stdin
Sending cancel signal to another thread
I am in pthread handler

--
Thanks in advance
>
>
>
9/15/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish:

From what you've written, if you're doing a blocking read(2) call, and

a cancellation is posted, it would deliver it and like you said,
read(2) will return and reflect EINTR in the errno. BUT, if its just
executing
read(2) call, but not blocked, it would depend on whether cancellation

is delivered to it or not, since it would deliver to it only in
certain states.

cancellation is delivered, testcancel will not catch it. Could
you share the code snippet that you're trying, with us?

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish
Katiyar
Sent: Friday, September 15, 2006 11:45 AM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Dennnis/Rohit,

Thanks for your responsesCould you please explain me how does
HPUX pthread_cancel works in case of blocking system calls. EgIf i
am in middle of read system call and get a cancellation signal. read
will return with EINTR am i right ?and after that cancel
signal still exists or is said to be deliveredcoz If i do a
testcancel after that I can still see that signal
--
thanks in advance.
>
>
>
9/13/06, Manish Katiyar <mkatiyar (AT) gmail (DOT) comwrote:

Hi Dennis,

Thanks for the responseyes main thread should not be cancelled

but I see this behaviour when i create a thread alsojust to
keep the code short I cancelled the main thread.

"I assume that since you aren't calling any libc functions, there is

no place where synchronous cancels can be caught."

I partially agree with this coz as I know libc functions just act as

additional cancellation points. Even if they are not there and
thread behaviour is deferred, signal should be delivered to the
thread at some point and this is what I expect.

Any other suggestions ?
>
>
>

Hi Rohit,

Could you identify the reason for the behaviour of the program?Thanks in
advance

9/15/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish: Firstly, my apology for a stupid mistake that I made. I'd
introduced a perror() in the program, and it seems that perror was
catching the cancel request and not the read(). The behaviour you're
witnessing is the behaviour on 11.11 and 11.23 too.

Will respond shortly on why we're witnessing this behaviour. The
man-page seems to mention that read() "can be" a cancellation point but
Cancel-Safe and Async-cancel-safe sections do not list read() as an
interface under either of the categories.

Will get back.

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish Katiyar
Sent: Friday, September 15, 2006 5:36 PM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Rohit,

I am testing it on both 11.11 PA and 11.23 PAand on both it
doesn't work without compiling it with -DAIs there any patch
issueI have the patch 37418 applied on my 11.23 machine.

/home/katiyar./a.out

I have pushed a function
Going to block while read from stdin
Sending cancel signal to another thread
/home/katiyaruname -a
HP-UX hpsarpc1 B.11.11 U 9000/800 1579700508 unlimited-user license
/home/katiyar>
--
/tmp/katiyarcc -o HP -lpthread HP.c
/tmp/katiyar./HP

I have pushed a function
Going to block while read from stdin
Sending cancel signal to another thread
/tmp/katiyaruname -a
HP-UX holmium B.11.23 U 9000/800 1759078586 unlimited-user license
>
>
>
Thanks in advance

9/15/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish: I've tried the program on both 11.11(PA) and 11.23 (IA)
servers.
Both the cases, the cleanup handler is called, with and without -DA
options.

Also, a small correction to what you'd stated in your mail previously
and I'd asserted that to be true - when you post the cancellation, the

thread doesn't return from the cancellation point, the thread is
terminated thereafter. If there is a registered cleanup handler, the
termination would be post the action of the handler.

The output I got on my machines is pasted below:

$ uname -a
HP-UX rocket B.11.11 U 9000/800 187494646 unlimited-user license $ cc
-o a -lpthread a.c $ ./a

I have pushed a function
Going to block while read from stdin
Sending cancel signal to another thread I am in pthread handler $ cc
-DA -o a -lpthread a.c $ ./a

I have pushed a function
Going to block while read from stdin
Sending cancel signal to another thread I am in pthread handler

Are you on 11.11(PA), 11.23 (IA) or 11.0?

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish
Katiyar
Sent: Friday, September 15, 2006 3:19 PM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Rohit,

Below is my code snippet and its output
--

#include<stdio.h>
#include<signal.h>
#include<pthread.h>

pthread_t stid,ftid;

void hndlr(void *a){
int oldstate;
printf("I am in pthread handler\n");
return ;
}
void *fstart(void *tid){
printf("Sending cancel signal to another thread \n");
pthread_cancel(stid);
return NULL;
}

void *sstart(void *tid){
char buff[100];
int n;
pthread_cleanup_push(hndlr,NULL);
printf("\nI have pushed a function\n");
printf("Going to block while read from stdin\n");
n=read(0,buff,100);
#ifdef A
pthread_testcancel();
#endif
pthread_cleanup_pop(0);
return NULL ;
}

int main(){
pthread_attr_t attr;
pthread_attr_init(&attr);
pthread_create(&stid,&attr,sstart,NULL);
sleep(3);
pthread_create(&ftid,&attr,fstart,NULL);
pthread_join(stid,NULL);
return 0;
}

/tmp/katiyarcc -o HP -lpthread HP.c
/tmp/katiyar./HP

I have pushed a function
Going to block while read from stdin
Sending cancel signal to another thread /tmp/katiyarcc -DA -o HP
-lpthread HP.c /tmp/katiyar./HP

I have pushed a function
Going to block while read from stdin
Sending cancel signal to another thread I am in pthread handler

--
Thanks in advance
>
>
>
9/15/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish:

From what you've written, if you're doing a blocking read(2) call,
and

a cancellation is posted, it would deliver it and like you said,
read(2) will return and reflect EINTR in the errno. BUT, if its just

executing
read(2) call, but not blocked, it would depend on whether
cancellation

is delivered to it or not, since it would deliver to it only in
certain states.

cancellation is delivered, testcancel will not catch it. Could
you share the code snippet that you're trying, with us?

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish
Katiyar
Sent: Friday, September 15, 2006 11:45 AM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Dennnis/Rohit,

Thanks for your responsesCould you please explain me how does
HPUX pthread_cancel works in case of blocking system calls. EgIf
i am in middle of read system call and get a cancellation signal.
read will return with EINTR am i right ?and after that
cancel signal still exists or is said to be deliveredcoz If i

do a testcancel after that I can still see that signal
--
thanks in advance.
>
>
>
9/13/06, Manish Katiyar <mkatiyar (AT) gmail (DOT) comwrote:

Hi Dennis,

Thanks for the responseyes main thread should not be
cancelled

but I see this behaviour when i create a thread alsojust to

keep the code short I cancelled the main thread.

"I assume that since you aren't calling any libc functions, there
is

no place where synchronous cancels can be caught."

I partially agree with this coz as I know libc functions just act
as

additional cancellation points. Even if they are not there and
thread behaviour is deferred, signal should be delivered to the
thread at some point and this is what I expect.

Any other suggestions ?
>
>
>

Thanks a lot Rohit for your time and advices.

9/18/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish: I got involved with something else. Please gimme sometime. I'll
reply today, certainly.

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish Katiyar
Sent: Monday, September 18, 2006 3:06 PM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Rohit,

Could you identify the reason for the behaviour of the
program?Thanks in advance

9/15/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish: Firstly, my apology for a stupid mistake that I made. I'd
introduced a perror() in the program, and it seems that perror was
catching the cancel request and not the read(). The behaviour you're
witnessing is the behaviour on 11.11 and 11.23 too.

Will respond shortly on why we're witnessing this behaviour. The
man-page seems to mention that read() "can be" a cancellation point
but Cancel-Safe and Async-cancel-safe sections do not list read() as
an interface under either of the categories.

Will get back.

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish
Katiyar
Sent: Friday, September 15, 2006 5:36 PM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Rohit,

I am testing it on both 11.11 PA and 11.23 PAand on both it
doesn't work without compiling it with -DAIs there any patch
issueI have the patch 37418 applied on my 11.23 machine.

/home/katiyar./a.out

I have pushed a function
Going to block while read from stdin
Sending cancel signal to another thread /home/katiyaruname -a HP-UX
hpsarpc1 B.11.11 U 9000/800 1579700508 unlimited-user license
/home/katiyar>
--
/tmp/katiyarcc -o HP -lpthread HP.c
/tmp/katiyar./HP

I have pushed a function
Going to block while read from stdin
Sending cancel signal to another thread /tmp/katiyaruname -a HP-UX
holmium B.11.23 U 9000/800 1759078586 unlimited-user license
>
>
>
Thanks in advance

9/15/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish: I've tried the program on both 11.11(PA) and 11.23 (IA)
servers.
Both the cases, the cleanup handler is called, with and without -DA
options.

Also, a small correction to what you'd stated in your mail
previously and I'd asserted that to be true - when you post the
cancellation, the

thread doesn't return from the cancellation point, the thread is
terminated thereafter. If there is a registered cleanup handler, the

termination would be post the action of the handler.

The output I got on my machines is pasted below:

$ uname -a
HP-UX rocket B.11.11 U 9000/800 187494646 unlimited-user license $
cc -o a -lpthread a.c $ ./a

I have pushed a function
Going to block while read from stdin Sending cancel signal to
another thread I am in pthread handler $ cc -DA -o a -lpthread a.c $

./a

I have pushed a function
Going to block while read from stdin Sending cancel signal to
another thread I am in pthread handler

Are you on 11.11(PA), 11.23 (IA) or 11.0?

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish
Katiyar
Sent: Friday, September 15, 2006 3:19 PM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Rohit,

Below is my code snippet and its output
--

#include<stdio.h>
#include<signal.h>
#include<pthread.h>

pthread_t stid,ftid;

void hndlr(void *a){
int oldstate;
printf("I am in pthread handler\n");
return ;
}
void *fstart(void *tid){
printf("Sending cancel signal to another thread \n");
pthread_cancel(stid);
return NULL;
}

void *sstart(void *tid){
char buff[100];
int n;
pthread_cleanup_push(hndlr,NULL);
printf("\nI have pushed a function\n");
printf("Going to block while read from stdin\n");
n=read(0,buff,100);
#ifdef A
pthread_testcancel();
#endif
pthread_cleanup_pop(0);
return NULL ;
}

int main(){
pthread_attr_t attr;
pthread_attr_init(&attr);
pthread_create(&stid,&attr,sstart,NULL);
sleep(3);
pthread_create(&ftid,&attr,fstart,NULL);
pthread_join(stid,NULL);
return 0;
}

/tmp/katiyarcc -o HP -lpthread HP.c /tmp/katiyar./HP

I have pushed a function
Going to block while read from stdin Sending cancel signal to
another thread /tmp/katiyarcc -DA -o HP -lpthread HP.c
/tmp/katiyar./HP

I have pushed a function
Going to block while read from stdin Sending cancel signal to
another thread I am in pthread handler

--
Thanks in advance
>
>
>
9/15/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish:

From what you've written, if you're doing a blocking read(2) call,

and

a cancellation is posted, it would deliver it and like you said,
read(2) will return and reflect EINTR in the errno. BUT, if its
just

executing
read(2) call, but not blocked, it would depend on whether
cancellation

is delivered to it or not, since it would deliver to it only in
certain states.

cancellation is delivered, testcancel will not catch it.
Could you share the code snippet that you're trying, with us?

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish
Katiyar
Sent: Friday, September 15, 2006 11:45 AM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Dennnis/Rohit,

Thanks for your responsesCould you please explain me how
does HPUX pthread_cancel works in case of blocking system calls.
EgIf i am in middle of read system call and get a cancellation
signal.
read will return with EINTR am i right ?and after that
cancel signal still exists or is said to be deliveredcoz If

i

do a testcancel after that I can still see that signal
--
thanks in advance.
>
>
>
9/13/06, Manish Katiyar <mkatiyar (AT) gmail (DOT) comwrote:

Hi Dennis,

Thanks for the responseyes main thread should not be
cancelled

but I see this behaviour when i create a thread alsojust
to

keep the code short I cancelled the main thread.

"I assume that since you aren't calling any libc functions,
there is

no place where synchronous cancels can be caught."

I partially agree with this coz as I know libc functions just
act as

additional cancellation points. Even if they are not there and
thread behaviour is deferred, signal should be delivered to the
thread at some point and this is what I expect.

Any other suggestions ?
>
>
>

Hi Rohit,
yes me too was getting the same behaviourthat is why I thought
that probably read call after getting the signal is re-raising
itprobably someone who is more familiar with internals of read can
help

Thanks in advance for any suggestions

9/18/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish: I see that for some reason, the read(2) system call is getting
restarted instead of being terminated with an EINTR. I tried using a
non-tty file stream but the behaviour persisted.

Will get back once I get more info on this. Till then, anyone else?

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish Katiyar
Sent: Monday, September 18, 2006 3:06 PM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Rohit,

Could you identify the reason for the behaviour of the
program?Thanks in advance

9/15/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish: Firstly, my apology for a stupid mistake that I made. I'd
introduced a perror() in the program, and it seems that perror was
catching the cancel request and not the read(). The behaviour you're
witnessing is the behaviour on 11.11 and 11.23 too.

Will respond shortly on why we're witnessing this behaviour. The
man-page seems to mention that read() "can be" a cancellation point
but Cancel-Safe and Async-cancel-safe sections do not list read() as
an interface under either of the categories.

Will get back.

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish
Katiyar
Sent: Friday, September 15, 2006 5:36 PM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Rohit,

I am testing it on both 11.11 PA and 11.23 PAand on both it
doesn't work without compiling it with -DAIs there any patch
issueI have the patch 37418 applied on my 11.23 machine.

/home/katiyar./a.out

I have pushed a function
Going to block while read from stdin
Sending cancel signal to another thread /home/katiyaruname -a HP-UX
hpsarpc1 B.11.11 U 9000/800 1579700508 unlimited-user license
/home/katiyar>
--
/tmp/katiyarcc -o HP -lpthread HP.c
/tmp/katiyar./HP

I have pushed a function
Going to block while read from stdin
Sending cancel signal to another thread /tmp/katiyaruname -a HP-UX
holmium B.11.23 U 9000/800 1759078586 unlimited-user license
>
>
>
Thanks in advance

9/15/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish: I've tried the program on both 11.11(PA) and 11.23 (IA)
servers.
Both the cases, the cleanup handler is called, with and without -DA
options.

Also, a small correction to what you'd stated in your mail
previously and I'd asserted that to be true - when you post the
cancellation, the

thread doesn't return from the cancellation point, the thread is
terminated thereafter. If there is a registered cleanup handler, the

termination would be post the action of the handler.

The output I got on my machines is pasted below:

$ uname -a
HP-UX rocket B.11.11 U 9000/800 187494646 unlimited-user license $
cc -o a -lpthread a.c $ ./a

I have pushed a function
Going to block while read from stdin Sending cancel signal to
another thread I am in pthread handler $ cc -DA -o a -lpthread a.c $

./a

I have pushed a function
Going to block while read from stdin Sending cancel signal to
another thread I am in pthread handler

Are you on 11.11(PA), 11.23 (IA) or 11.0?

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish
Katiyar
Sent: Friday, September 15, 2006 3:19 PM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Rohit,

Below is my code snippet and its output
--

#include<stdio.h>
#include<signal.h>
#include<pthread.h>

pthread_t stid,ftid;

void hndlr(void *a){
int oldstate;
printf("I am in pthread handler\n");
return ;
}
void *fstart(void *tid){
printf("Sending cancel signal to another thread \n");
pthread_cancel(stid);
return NULL;
}

void *sstart(void *tid){
char buff[100];
int n;
pthread_cleanup_push(hndlr,NULL);
printf("\nI have pushed a function\n");
printf("Going to block while read from stdin\n");
n=read(0,buff,100);
#ifdef A
pthread_testcancel();
#endif
pthread_cleanup_pop(0);
return NULL ;
}

int main(){
pthread_attr_t attr;
pthread_attr_init(&attr);
pthread_create(&stid,&attr,sstart,NULL);
sleep(3);
pthread_create(&ftid,&attr,fstart,NULL);
pthread_join(stid,NULL);
return 0;
}

/tmp/katiyarcc -o HP -lpthread HP.c /tmp/katiyar./HP

I have pushed a function
Going to block while read from stdin Sending cancel signal to
another thread /tmp/katiyarcc -DA -o HP -lpthread HP.c
/tmp/katiyar./HP

I have pushed a function
Going to block while read from stdin Sending cancel signal to
another thread I am in pthread handler

--
Thanks in advance
>
>
>
9/15/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish:

From what you've written, if you're doing a blocking read(2) call,

and

a cancellation is posted, it would deliver it and like you said,
read(2) will return and reflect EINTR in the errno. BUT, if its
just

executing
read(2) call, but not blocked, it would depend on whether
cancellation

is delivered to it or not, since it would deliver to it only in
certain states.

cancellation is delivered, testcancel will not catch it.
Could you share the code snippet that you're trying, with us?

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish
Katiyar
Sent: Friday, September 15, 2006 11:45 AM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Dennnis/Rohit,

Thanks for your responsesCould you please explain me how
does HPUX pthread_cancel works in case of blocking system calls.
EgIf i am in middle of read system call and get a cancellation
signal.
read will return with EINTR am i right ?and after that
cancel signal still exists or is said to be deliveredcoz If

i

do a testcancel after that I can still see that signal
--
thanks in advance.
>
>
>
9/13/06, Manish Katiyar <mkatiyar (AT) gmail (DOT) comwrote:

Hi Dennis,

Thanks for the responseyes main thread should not be
cancelled

but I see this behaviour when i create a thread alsojust
to

keep the code short I cancelled the main thread.

"I assume that since you aren't calling any libc functions,
there is

no place where synchronous cancels can be caught."

I partially agree with this coz as I know libc functions just
act as

additional cancellation points. Even if they are not there and
thread behaviour is deferred, signal should be delivered to the
thread at some point and this is what I expect.

Any other suggestions ?
>
>
>

Thanks Rohit,

for your efforts. I will open a bug with HP regarding this. Has HP any
official site as Linux's bugzilla for this ?I could not find any link
where I can report a bug to HP from google or forums.itrc.hp.com.

Thanks again.

9/22/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish: The behaviour seems to be a bug in read(2) system call. Could I
request you to file a bug report on the same.

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish Katiyar
Sent: Tuesday, September 19, 2006 8:18 AM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Rohit,
yes me too was getting the same behaviourthat is why I thought
that probably read call after getting the signal is re-raising
itprobably someone who is more familiar with internals of read
can help

Thanks in advance for any suggestions

9/18/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish: I see that for some reason, the read(2) system call is getting

restarted instead of being terminated with an EINTR. I tried using a
non-tty file stream but the behaviour persisted.

Will get back once I get more info on this. Till then, anyone else?

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish
Katiyar
Sent: Monday, September 18, 2006 3:06 PM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Rohit,

Could you identify the reason for the behaviour of the
program?Thanks in advance

9/15/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish: Firstly, my apology for a stupid mistake that I made. I'd
introduced a perror() in the program, and it seems that perror was
catching the cancel request and not the read(). The behaviour you're

witnessing is the behaviour on 11.11 and 11.23 too.

Will respond shortly on why we're witnessing this behaviour. The
man-page seems to mention that read() "can be" a cancellation point
but Cancel-Safe and Async-cancel-safe sections do not list read() as

an interface under either of the categories.

Will get back.

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish
Katiyar
Sent: Friday, September 15, 2006 5:36 PM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Rohit,

I am testing it on both 11.11 PA and 11.23 PAand on both it

doesn't work without compiling it with -DAIs there any patch
issueI have the patch 37418 applied on my 11.23 machine.

/home/katiyar./a.out

I have pushed a function
Going to block while read from stdin Sending cancel signal to
another thread /home/katiyaruname -a HP-UX
hpsarpc1 B.11.11 U 9000/800 1579700508 unlimited-user license
/home/katiyar>
--
/tmp/katiyarcc -o HP -lpthread HP.c /tmp/katiyar./HP

I have pushed a function
Going to block while read from stdin Sending cancel signal to
another thread /tmp/katiyaruname -a HP-UX holmium B.11.23 U
9000/800 1759078586 unlimited-user license
>
>
>
Thanks in advance

9/15/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish: I've tried the program on both 11.11(PA) and 11.23 (IA)
servers.
Both the cases, the cleanup handler is called, with and without
-DA options.

Also, a small correction to what you'd stated in your mail
previously and I'd asserted that to be true - when you post the
cancellation, the

thread doesn't return from the cancellation point, the thread is
terminated thereafter. If there is a registered cleanup handler,
the

termination would be post the action of the handler.

The output I got on my machines is pasted below:

$ uname -a
HP-UX rocket B.11.11 U 9000/800 187494646 unlimited-user license $

cc -o a -lpthread a.c $ ./a

I have pushed a function
Going to block while read from stdin Sending cancel signal to
another thread I am in pthread handler $ cc -DA -o a -lpthread a.c

$

./a

I have pushed a function
Going to block while read from stdin Sending cancel signal to
another thread I am in pthread handler

Are you on 11.11(PA), 11.23 (IA) or 11.0?

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish
Katiyar
Sent: Friday, September 15, 2006 3:19 PM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Rohit,

Below is my code snippet and its output
--

#include<stdio.h>
#include<signal.h>
#include<pthread.h>

pthread_t stid,ftid;

void hndlr(void *a){
int oldstate;
printf("I am in pthread handler\n");
return ;
}
void *fstart(void *tid){
printf("Sending cancel signal to another thread \n");
pthread_cancel(stid);
return NULL;
}

void *sstart(void *tid){
char buff[100];
int n;
pthread_cleanup_push(hndlr,NULL);
printf("\nI have pushed a function\n");
printf("Going to block while read from stdin\n");
n=read(0,buff,100);
#ifdef A
pthread_testcancel();
#endif
pthread_cleanup_pop(0);
return NULL ;
}

int main(){
pthread_attr_t attr;
pthread_attr_init(&attr);
pthread_create(&stid,&attr,sstart,NULL);
sleep(3);
pthread_create(&ftid,&attr,fstart,NULL);
pthread_join(stid,NULL);
return 0;
}

/tmp/katiyarcc -o HP -lpthread HP.c /tmp/katiyar./HP

I have pushed a function
Going to block while read from stdin Sending cancel signal to
another thread /tmp/katiyarcc -DA -o HP -lpthread HP.c
/tmp/katiyar./HP

I have pushed a function
Going to block while read from stdin Sending cancel signal to
another thread I am in pthread handler

--
Thanks in advance
>
>
>
9/15/06, Gupta, Rohit (STSD) <rohit.gupta (AT) hp (DOT) comwrote:

Manish:

From what you've written, if you're doing a blocking read(2)
call,

and

a cancellation is posted, it would deliver it and like you said,
read(2) will return and reflect EINTR in the errno. BUT, if its
just

executing
read(2) call, but not blocked, it would depend on whether
cancellation

is delivered to it or not, since it would deliver to it only in
certain states.

cancellation is delivered, testcancel will not catch it.
Could you share the code snippet that you're trying, with us?

Regards,
Rohit

Message
From: owner-hpux-devtools (AT) cxx (DOT) cup.hp.com
[mailto:owner-hpux-devtools (AT) cxx (DOT) cup.hp.com] Behalf Manish
Katiyar
Sent: Friday, September 15, 2006 11:45 AM
To: hpux-devtools (AT) cxx (DOT) cup.hp.com
Subject: Re: HPUX-DEVTLS: pthread_cancel doesn't work

Hi Dennnis/Rohit,

Thanks for your responsesCould you please explain me how
does HPUX pthread_cancel works in case of blocking system calls.
EgIf i am in middle of read system call and get a
cancellation
signal.
read will return with EINTR am i right ?and after
that cancel signal still exists or is said to be
deliveredcoz If

i

do a testcancel after that I can still see that signal
--
thanks in advance.
>
>
>
9/13/06, Manish Katiyar <mkatiyar (AT) gmail (DOT) comwrote:

Hi Dennis,

Thanks for the responseyes main thread should not be
cancelled

but I see this behaviour when i create a thread
alsojust to

keep the code short I cancelled the main thread.

"I assume that since you aren't calling any libc functions,
there is

no place where synchronous cancels can be caught."

I partially agree with this coz as I know libc functions just
act as

additional cancellation points. Even if they are not there and

thread behaviour is deferred, signal should be delivered to
the thread at some point and this is what I expect.

Any other suggestions ?
>
>
>