Re: [docs] "FILES" glossary entry does not mention including "dirname/*"

Hi Robert,

On 5/29/26 1:00 PM, Robert P. J. Day via lists.yoctoproject.org wrote:
> 
>    explaining to a colleague how to properly define a package's
> "FILES:${PN}" value, and noticed that the explanation here:
> 
>    https://docs.yoctoproject.org/ref-manual/variables.html#term-FILES
> 
> has a subtle distinction from the default value in bitbake.conf. in
> the glossary, the sample setting:
> 
>    FILES:${PN} += "${bindir}/mydir1 ${bindir}/mydir2/myfile"
> 
> shows how to include an entire directory -- just list the directory
> name and the inclusion is recursive.
> 
>    my colleague had initially used the notation "dirname/*", and i
> pointed out that the "*" wildcard was superfluous since, you know,
> recursive, but he showed me the sample he had used from bitbake.conf,
> which *does* include that wildcard:
> 
>    FILES:${PN} = "${bindir}/* ${sbindir}/* ${libexecdir}/* ...
> 
> i couldn't immediately explain why bitbake.conf used the wildcard, as
> the only difference i could imagine is that explicitly using the "*"
> wildcard would ignore hidden paths, but that seemed like a weak
> explanation.
> 
>    is there some subtlety i'm missing?
> 

Did you ever figure this one out?

I started to look into it quickly because I vaguely remember some 
gotchas around including directories and symlinks we ran into at my 
previous company years ago.

I'm assuming you want to have a look at populate_packages() in 
meta/lib/oe/package.py.
It'll eventually call oe.package.files_from_filevars() which runs 
glob.glob(path, recursive=True) on each path. The issue is that:

 >>> import glob
 >>> glob.glob('./tools/', recursive=True)
['./tools/']
 >>> glob.glob('./tools', recursive=True)
['./tools']
 >>> glob.glob('./tools/*', recursive=True)
['./tools/accounting', './tools/arch', './tools/bootconfig', 
'./tools/bpf', './tools/build', './tools/cgroup', './tools/counter', 
'./tools/debugging', './tools/firewire', './tools/firmware', 
'./tools/gpio', './tools/hv', './tools/iio', './tools/include', 
'./tools/kvm', './tools/laptop', './tools/leds', './tools/lib', 
'./tools/memory-model', './tools/objtool', './tools/pci', 
'./tools/pcmcia', './tools/perf', './tools/power', './tools/rcu', 
'./tools/scripts', './tools/spi', './tools/testing', './tools/thermal', 
'./tools/time', './tools/tracing', './tools/usb', './tools/virtio', 
'./tools/wmi', './tools/Makefile', './tools/certs', './tools/crypto', 
'./tools/mm', './tools/net', './tools/sched_ext', './tools/sound', 
'./tools/verification', './tools/workqueue', './tools/writeback']

and I cannot make sense of the rest of the logic in populate_packages() 
which checks whether the path is a directory, and if so, seemingly only 
runs a homegrown version of mkdir -p. Something's not adding up here.

Note that glob.glob() doesn't return hidden files by default, c.f. 
https://docs.python.org/3/library/glob.html#glob.glob (feature only 
added in Python 3.11 anyway, which we haven't set as our minimally 
supported version).

Cheers,
Quentin