On Sun, 21 Feb 2021 18:54:01 +0300, Andrew Rybchenko wrote:

On 2/21/21 11:59 AM, Tal Shnaiderman wrote:

quoted

quoted

Subject: [dpdk-dev] [PATCH v2 6/7] drivers: remove POSIX dependencies

External email: Use caution opening links or attachments


Replace POSIX strncasecmp() with EAL rte_strncasecmp().
Replace POSIX strtok_r() with EAL rte_strtok().
Replace POSIX strdup() with EAL rte_strdup().

Signed-off-by: Dmitry Kozlyuk <redacted>
---
i40e: checkpatches.sh complains about long lines (it's ~85).
       I doubt that mechanical fix would keep the code readable.
       It's on 5th level of indentation, so I'd extract a function,
       but would like to hear from maintainers first.

  drivers/bus/pci/private.h             |  2 +-
  drivers/bus/vdev/vdev.c               |  4 +-
  drivers/bus/vdev/vdev_params.c        |  3 +-
  drivers/common/mlx5/mlx5_common_pci.c |  4 +-  

bus_cmdline_options_handler in mlx5_common_pci.c has a call to strdup which needs to be renamed to rte_strdup
(Also failed CI: https://lab.dpdk.org/results/dashboard/patchsets/15674/ )
  

Frankly speaking I don't understand why such changes are useful/needed.
Patch description does not explain/prove it.

Please see the cover letter (better to partially repeat in each patch?):

	https://mails.dpdk.org/archives/dev/2021-February/199981.html

Driver code using POSIX functions is not really portable without a shim.
Having a shim in public API is wrong, it makes DPDK expose non-DPDK symbols.

An alternative approach would be to make the shim DPDK-internal. It would
require just a change in #includes for drivers and libs. My doubt is whether
EAL should emulate POSIX.