[aur-general] rfc: pkgbuild for prospect releng-tool

James Knight james.d.knight at live.com
Thu Mar 7 02:53:24 UTC 2019


On 2019-03-06 1:59 AM, Doug Newgard via aur-general wrote:
> On Tue, 5 Mar 2019 23:53:10 -0700
> Brett Cornwall via aur-general <aur-general at archlinux.org> wrote:
>> Capitalize the T!

Can do!

>> You're using python-setuptools, so you'll want to set that in
>> makedepends instead of python.

Noted.

> Also missing git as a makedepends. Might want to build in a clean chroot and see
> what else you're missing. python shouldn't be in the makedepends, as it should be
> in the global deps array, as stated below.

Based off a comment in a related thread, I may be opting to switching 
away from the VCS type to just an archive. If for some reason I still 
decide to stick with Git, I will make sure the dependency is there.

I will try to perform additional checks to make sure I have the required 
dependencies listed.

>>> package() {
>>>   depends=('python')
>>
>> The depends() should just go to the top alongside makedepends for this
>> package. You probably saw this in the examples for the python packaging
>> standards, but this is typically used for a 'split package', i.e. using
>> one PKGBUILD to build versions for both python2 and python3. Since
>> you're only building for python 3 depends() should go to the top.

Thanks. Based off some comments in a related thread, I may go back to my 
initial attempt in supporting split-packages (as the tool supports 
running on both Python major versions). I will try to make sure I 
cleanup the dependencies on next pass. Truthfully, I did not even notice 
it said "makedepends"; I just blindly assumed it was the way to list 
dependencies. Time to re-read some more documentation.
>>>   python setup.py install --root="$pkgdir" --optimize=1
>>
>> Go ahead and add a --skip-build here since you already built earlier.

Odd, I swear I added that. Good catch.

>>
>> [...]
>>
>>>   install -dm 755 "$pkgdir/usr/share/bash-completion/completions"
>>>   install -m644 scripts/releng-tool-completion
>>> "$pkgdir/usr/share/bash-completion/completions/releng-tool"
>>
>> No need to create the directory beforehand; This can be shortened into:
>>
>>      install -Dm644 scripts/releng-tool-completion "$pkgdir/usr/share/bash-completion/completions/releng-tool"

I recall seeing another PKGBUILD definition splitting the install 
operations in two-parts (although I cannot seem to find it at this 
time). I assume it was to prevent the risk of a host adjusting their 
umask value and producing unexpected results (i.e. the way submitted 
ensures directories are built with "755" permissions and the install 
script is installed with "644" permissions. Although this scrutiny falls 
apart when the submitted PKGBUILD performs the license installation with 
a care to ensure an explicit permission set for any generated folders.

I do not mind simplifying the change into a single line. I am just 
curious why this is not an issue (maybe I am overlooking something)?

---

The comments are appreciated.

On a thread note, I am curious to know why I only saw a single Email for 
this response, which clearly shows two individuals comments. My 
curiosity is only driven by my concern to ensure I am receiving Emails 
properly. If what I receive is an expected output from the mailing list, 
this comment can be disregarded.


More information about the aur-general mailing list