Re: ci: version.map check

[Ferruh Yigit <ferruh.yigit@amd.com>]

On 9/11/2024 1:08 AM, Morten Brørup wrote:
>> From: Ferruh Yigit [mailto:ferruh.yigit@amd.com]
>> Sent: Wednesday, 11 September 2024 01.48
>>
>> On 9/10/2024 5:56 PM, Morten Brørup wrote:
>>> It would be convenient if the CI checked that the symbols in
>> version.map files actually exist in the resulting code.
>>>
>>
>> Is there instances that is not the case?
> 
> While reviewing a patch, I noticed one such instance.
> A previous version of the patch contained a global variable that had since been removed, but not from the version.map file.
> 
> Specifically "rte_lcore_var" in this earlier version:
> https://patchwork.dpdk.org/project/dpdk/patch/20240208181644.455233-2-mattias.ronnblom@ericsson.com/
> 
> Which in this more recent patch has been removed, but remains in version.map:
> https://patchwork.dpdk.org/project/dpdk/patch/20240910070344.699183-2-mattias.ronnblom@ericsson.com/
> 

I expect build to catch them, via `./buildtools/check-symbols.sh`, which
uses `./devtools/check-symbol-maps.sh` internally. I did a quick check,
and add a non-existing symbol to a version.map file, build failed [1] as
expected.

For 'rte_lcore_var' instance, `check-symbol-maps.sh` does not catch it,
but if you update the variable name in version.map to 'rte_lcore_bar' it
does catch :)

So I assume `check-symbol-maps.sh` uses wildcard search, and because of
existing 'rte_lcore_var_alloc', detection fails.

Probably we need to fix `check-symbol-maps.sh` for this, and build will
catch extra symbols.


[1]
[133/264] Generating lib/ethdev.sym_chk with a custom command (wrapped
by meson to capture output)
FAILED: lib/ethdev.sym_chk