Process substitution with bash 5.2.0 (Build 551)

[gngrossi]

Process substitution with bash 5.2.0 (Build 551)

[gngrossi]

GNU bash, version 5.2.0(1)-release (i370-ibm-openedition)

RC=(0) [SYSA] bash-5.2$ cat <(date) | xxd
00000000: e8cd c180 282f ca80 9194 8091 969a 9590 ▒▒▒..ʀ▒▒▒▒▒▒▒▒
00000010: 9a90 9180 e4e0 e880 1690 1693 8e ▒▒▒▒▒▒▒.▒.▒▒

RC=(0) [SYSA] bash-5.2$ cat <(date) | od -hc
0000000000 E8 CD C1 80 28 2F CA 80 91 94 80 91 96 9A 95 90
Y 315 A 200 050 \a 312 200 j m 200 j o 232 n 220
0000000020 9A 90 04 80 E4 E0 E8 80 16 90 16 93 8E
232 220 004 200 U \ Y 200 \b 220 \b l 216
0000000035

[gngrossi]

bashport_768/bash-5.2.20230319_200518.zos.pax.Z

This is now working as expected

bash-5.2$ cat <(date)
Mon Mar 20 18:18:27 CDT 2023

Is this working as expected? Should there be a /dev/fdxx returned?

bash-5.2$ echo <(date)
/u/@02858/tmp/sh-np.CechxQ

Getting the correct line count. Should there be a /dev/fdxx returned?

bash-5.2$ wc -l <(ssh l98misdb491 "cat *Yum")
4 /u/@02858/tmp/sh-np.AEBbxQ

[IgorTodorovskiIBM]

I think there's two things going on here.

1. It appears that our CI machines are not consistently setting HAVE_DEV_FD, not setting this triggers an alternate path to be used which create temporary files using mktemps. This is determine by the check test -r /dev/fd/0 which returns 0 on some of our CI machines and 1 on others.
2. zoslib does not appear to handle untagged character devices properly, such as /dev/fd*. I'll look into this

[gngrossi]

bash-5.2$ bash --version | head -n1
GNU bash, version 5.2.15(1)-release (i370-ibm-openedition)

bash-5.2$ cat <(date)
cat: /dev/fd/63: EDC5135I Not a directory.

looks good

bash-5.2$ echo <(date)
/dev/fd/63

bash-5.2$ wc -l <(ssh l98misdb491 "cat *Yum")
wc: /dev/fd/63: EDC5135I Not a directory.

[IgorTodorovskiIBM]

We noticed there's an inconsistency on some of our z/OS machines /dev/fd is a character device and not a directory. So we made a change to use FIFO special files and to also tag them as such:

[ITODORO@ZOSCAN2B ~/zopen/prod/bash/bin]$  ls -lT <(date)
t ISO8859-1   T=on  prw-------   1 ITODORO  SUDOGRP        0 Mar 24 11:28 /tmp/sh-np.ghEFxQ
[ITODORO@ZOSCAN2B ~/zopen/prod/bash/bin]$ /home/itodoro/zopen/prod/coreutils/bin/cat <(date)
Fri Mar 24 11:28:48 EDT 2023
[ITODORO@ZOSCAN2B ~/zopen/prod/bash/bin]$ /bin/cat  <(date)
Fri Mar 24 11:28:55 EDT 2023

[gngrossi]

Are these the final results to be expected? If so, I will close.
thanks

bash 5.2.15 (Build 799)

RC=(0) [SYSA] bash-5.2$ cat <(date)
Fri Mar 24 11:34:28 CDT 2023

RC=(0) [SYSA] bash-5.2$ echo <(date)
/u/@02858/tmp/sh-np.hcabxQ

RC=(0) [SYSA] bash-5.2$ wc -l <(ssh l98misdb491 "cat *Yum")
4 /u/@02858/tmp/sh-np.GAfdxQ

[IgorTodorovskiIBM]

Yes, for now at least. At some point I'll revisit the /dev/fd issue as it may result in better performance. Opened issue #59