26/01/2021 16:42, Bruce Richardson:

On Tue, Jan 26, 2021 at 04:31:36PM +0100, Thomas Monjalon wrote:

quoted

26/01/2021 15:39, Bruce Richardson:

quoted

Removing the ALLOW_INTERNAL_API is probably a good idea, but it does indeed
throw up the errors with clang - but not gcc, which is strange. The
offending headers seem to be (initially):

* rte_ethdev_vdev.h
* rte_ethdev_pci.h

Are these public header files, or should they skip header checking - and
installation - as internal-only?

They are helpers for the drivers, so should be internal.

Just to confirm, by that you mean that they should not be installed for
end-application use?

I think yes they are not needed to be installed.

In which case, I believe that "rte_ethdev_driver", "rte_ethdev_vdev" and
"rte_ethdev_pci" should be removed from the "headers" list in
"librte_ethdev/meson.build". Current file list there is:

headers = files('rte_ethdev.h',
        'rte_ethdev_driver.h',
        'rte_ethdev_pci.h',
        'rte_ethdev_trace.h',
        'rte_ethdev_trace_fp.h',
        'rte_ethdev_vdev.h',
        'rte_dev_info.h',
        'rte_flow.h',
        'rte_flow_driver.h',
        'rte_mtr.h',
        'rte_mtr_driver.h',
        'rte_tm.h',
        'rte_tm_driver.h')
headers_no_chkincs += files('rte_eth_ctrl.h',
        'rte_ethdev_core.h')

[Note that the "headers_no_chkincs" files still get installed, they just
skip header checking as indirect includes.]
Anything else that should be removed.

Also, tangential to this, it would probably be good if we could come up
with a different naming scheme for internal-only headers, to make the
difference between what is to be installed or not, a lot clearer!

The difference is supposed to be the rte_ prefix in the file name.
Doxygen will parse only these files:
	FILE_PATTERNS = rte_*.h cmdline.h

I agree we should fix the internal filenames and not install them.