On 13/06/2024 12.42, Ulrich Mueller wrote:
>>>>>> On Thu, 13 Jun 2024, Florian Schmaus wrote:
>>>> +_GREADME_DOC_DIR="usr/share/doc/${PF}"
>>> It is somewhat unusual to call insinto or docompress with a relative
>>> path. I'd use "/usr/share/doc/${PF}" here.
>>>
>>>> +_GREADME_REL_PATH="${_GREADME_DOC_DIR}/${_GREADME_FILENAME}"
>>> Why must this be a relative path? AFAICS it could be an absolute
>>> path in
>>> everywhere it is used. (You even add an explicit slash in places like
>>> ${ED}/${_GREADME_REL_PATH}).
> 
>> My idea was to denote relative paths by not including an initial slash
>> (/). This allows to write "${ED}/${_GREADME_REL_PATH}" without a
>> duplicate slash in the middle. I find
> 
>> ${ED}/${_GREADME_REL_PATH}"
> 
>> more readable when compared to
> 
>> ${ED}${_GREADME_REL_PATH}"
> 
> I think the variable would be renamed in that case, i.e.
> ${ED}${_GREADME_PATH} or ${ED}${_GREADME_ABS_PATH}.
> 
>> which seems to be in-line with
>> https://mgorny.pl/articles/the-ultimate-guide-to-eapi-7.html#d-ed-root-eroot-no-longer-have-a-trailing-slash
> 
> This has examples like ${D}${EPREFIX}/usr/share/foo: All path variables
> start with a slash, _don't_ end with a slash, and no slash between them
> when concatenating.
> 
>> But maybe I misunderstand what you are suggesting. You want
> 
>> _GREADME_DOC_DIR="/usr/share/doc/${PF}"
> 
>> instead of
> 
>> _GREADME_DOC_DIR="usr/share/doc/${PF}"
> 
>> right?
> 
> Correct.
> 
>> Then I'd also have to change to ${ED}${_GREADME_REL_PATH}", to avoid
>> the double slash. :(
> 
> I see no problem with this.

Will change accordingly then.


>>>> +	while read -r line; do elog "${line}"; done < "${EROOT}/${_GREADME_REL_PATH}"
>>> It is not guaranteed that the file exists in ${EROOT} at this point.
>>> See FEATURES="nodoc" in make.conf(5).
> 
>> Fair point. Fixed.
> 
> Well, it still won't display anything with FEATURES=nodoc. IIUC
> readme.gentoo-r1.eclass works around the problem by saving the file
> contents in an environment variable. (However, I see the problem that
> even then you couldn't compare old vs new file contents.)

I'd assume this is readme.gentoo-r1.eclass working around the readme 
file being compressed, and by doing so, coincidentally being able to 
display the readme even if FEATURES=nodoc.

It appears we have now two options:
A) Just like readme.gentoo-r1, store the content in an environment
    variable , to be able to show the content (unconditionally) in case
    of FEATURES=nodoc
B) Live with the fact that we will not be able to show the content on
    FEATURES=nodoc (but document this accordingly).

I have no strong opinion, therefore I am happy about hearing peoples 
thoughts on this.

- Flow