Re: EINTR

[Bruno Haible]

Hi,

More details about this:
> 3) On MacOS X, SIGSTOP and SIGCONT make not only read() fail with EINTR, but
>    fread() as well.
> 
>    Test case: [3]. Run it, type Hello<Return> then Ctrl-Z then fg. Result is:
> 
>    $ ./a.out
>    Hello
>    ^Z
>    [1]+  Stopped                 ./a.out
>    $ fg
>    ./a.out
>    fread: Interrupted system call
>    Hello
> 
>    $ ./a.out --unbuffered
>    Hello
>    Hello
>    ^Z
>    [1]+  Stopped                 ./a.out --unbuffered
>    $ fg
>    ./a.out --unbuffered
>    fread: Interrupted system call
> 
>    This was also reported in [4].
> 
>    I would expect that EINTR also occurs in write() and fwrite() when writing
>    to a pipe that is temporarily full, but I can't easily reproduce this.

Looking at the MacOS X 10.5 kernel, EINTR appears to be generated in these
situations:
  - read, write, ioctl only on tty devices,
  - select, poll,
  - sleep,
  - pthread_cond_*,
  - connect.

This explaing why I cannot reproduce a malfunction with input from a pipe, only
from a tty.

Let me try a couple of coreutils programs that use fread():

1) base64: affected

$ ./base64
Hello
^Z
[1]+  Stopped                 ./base64
$ fg
./base64
SGVsbG8K
./base64: read error: Interrupted system call

2) cksum: affected

$ ./cksum 
Hello
^Z
[1]+  Stopped                 ./cksum
$ fg
./cksum
World
./cksum: -: Interrupted system call

3) sha1sum: affected

$ ./sha1sum
Hello
^Z
[1]+  Stopped                 ./sha1sum
$ fg
./sha1sum
./sha1sum: -: Interrupted system call

4) od: affected

$ ./od -x
Hello
^Z
[1]+  Stopped                 ./od -x
$ fg
./od -x
./od: standard input: read error: Interrupted system call
0000000 6548 6c6c 0a6f
0000006

5) sort: affected

$ sort
Hello
^Z
[1]+  Stopped                 sort
$ fg
sort
sort: read failed: -: Interrupted system call


So indeed, many programs are affected. I think an fread() wrapper would be
appropriate. Shall we call it safe_fread or fread?

Bruno