Re: [PATCH 2/3] module: add Error arguments to module_load_one and module_load_qom_one

[Claudio Fontana <cfontana@suse.de>]

On 9/7/22 09:36, Gerd Hoffmann wrote:
>   Hi,
>  
>> For this module_load_qom_all() maybe Gerd has a bit more context on
>> was should be the error reporting here?
> 
> Use case for module_load_qom_all() is someone enumerating the qom
> objects available.  So we load all modules known to have all object
> types registered and can return a complete list.
> 
> It could be that some of the known modules are not there.  Consider a
> distro packaging modules which depend on shared libraries into optional
> sub-rpms, to reduce the dependency chain of core qemu.  So, with core
> qemu installed and (some of) the sub-rpms not installed
> module_load_qom_all() will obviously fail to load some modules.
> 
> But I don't think those errors should be reported.  The object types
> implemented by the missing modules will also be missing from the object
> type list ...
> 
> Example: hw-usb-host.so is not installed.
> 
>   => 'qemu -device help' should IMHO not report the module load error
>      and just not list the 'usb-host' device.
>   => 'qemu -device usb-host' should report the module load error.
> 
> take care,
>   Gerd
> 

Hi Gerd,

the thing is, we can distinguish between a module not being present (ENOENT, ENOTDIR),
from a module being present, but failing to load.

So the "module not there" thing does not need to be treated separately, because no warning/error will be emitted if the module is not there.

It is up to the user/caller to decide what to do with the condition "module not there", error out and quit, continue on, etc.

Thanks this helped,

Claudio