[bug#51179] gnu: Add yt-dlp.

* gnu/packages/video.scm (yt-dlp): New variable.

Suggested-by: bdju <bdju@tilde.team>
---
 gnu/packages/video.scm | 71 ++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 71 insertions(+)

Hi Robin,

Am Mittwoch, den 13.10.2021, 05:44 -0400 schrieb Robin Templeton:
> * gnu/packages/video.scm (yt-dlp): New variable.
> 
> Suggested-by: bdju <bdju@tilde.team>
> ---
>  gnu/packages/video.scm | 71
> ++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 71 insertions(+)
> 
> diff --git a/gnu/packages/video.scm b/gnu/packages/video.scm
> index fcac369f60..1aceac25eb 100644
> --- a/gnu/packages/video.scm
> +++ b/gnu/packages/video.scm
> @@ -2360,6 +2360,77 @@ YouTube.com and many more sites.")
>      (home-page "https://yt-dl.org")
>      (license license:public-domain)))
>  
> +(define-public yt-dlp
> +  (package
> +    (name "yt-dlp")
> +    (version "2021.10.10")
I think yt-dlp can (and ought to) inherit from youtube-dl, which might
simplify some of the below.
> +    (source (origin
> +              (method url-fetch)
> +              (uri (string-append "https://github.com/yt-dlp/yt-dlp/
> "
> +                                  "releases/download/"
> +                                  version "/yt-dlp.tar.gz"))
> +              (sha256
> +               (base32
> +                "1ywld4qhvsik970gbac1h3kvxb74r7150m5axq9r5nffdw5sz3v
> d"))
> +              (snippet
> +               '(begin
> +                  ;; Delete the pre-generated files, except for the
> man page
> +                  ;; which requires 'pandoc' to build.
> +                  (for-each delete-file '("yt-dlp"
> +                                          ;;pandoc is needed to
> generate
> +                                          ;;"yt-dlp.1"
> +                                          "completions/bash/yt-dlp"
> +                                          "completions/fish/yt-
> dlp.fish"
> +                                          "completions/zsh/_yt-
> dlp"))
> +                  #t))))
Is this the same as for youtube-dl?  If not, we might want to give
pandoc as native-input.  That will increase build times, but it ought
not to increase closure size.
> +    (build-system python-build-system)
> +    (arguments
> +     ;; The problem here is that the directory for the man page and
> completion
> +     ;; files is relative, and for some reason, setup.py uses the
> +     ;; auto-detected sys.prefix instead of the user-defined "
> --prefix=FOO".
> +     ;; So, we need pass the prefix directly.
> +     `(#:tests? #f ; Many tests fail. The test suite can be run with
> pytest.
> +       #:phases (modify-phases %standard-phases
> +                  (add-after 'unpack 'default-to-the-ffmpeg-input
> +                    (lambda _
> +                      ;; See <https://issues.guix.gnu.org/43418#5>;.
> +                      ;; ffmpeg is big but required to request free
> formats
> +                      ;; from, e.g., YouTube so pull it in
> unconditionally.
> +                      ;; Continue respecting the --ffmpeg-location
> argument.
> +                      (substitute* "yt_dlp/postprocessor/ffmpeg.py"
> +                        (("\\.get\\('ffmpeg_location'\\)" match)
> +                         (format #f "~a or '~a'" match (which
> "ffmpeg"))))
> +                      #t))
> +                  (add-before 'build 'build-generated-files
> +                    (lambda _
> +                      ;; Avoid the make targets that require pandoc.
> +                      (invoke "make"
> +                              "PYTHON=python"
> +                              "yt-dlp"
> +                              ;;"youtube-dl.1"   ; needs pandoc
> +                              "completions")))
> +                  (add-before 'install 'fix-the-data-directories
> +                    (lambda* (#:key outputs #:allow-other-keys)
> +                      (let ((prefix (assoc-ref outputs "out")))
> +                        (substitute* "setup.py"
> +                          (("'etc/")
> +                           (string-append "'" prefix "/etc/"))
> +                          (("'share/")
> +                           (string-append "'" prefix "/share/")))
> +                        #t))))))
Horizontal space is at a premium and you can save some.
> +    (native-inputs
> +     `(("zip" ,zip)))
> +    (inputs
> +     `(("ffmpeg" ,ffmpeg)))
> +    (synopsis "Download videos from YouTube.com and other sites")
> +    (description
> +     "yt-dlp is a small command-line program to download videos from
> +YouTube.com and many more sites.  It is a fork of youtube-dl with a
> +focus on adding new features while keeping up-to-date with the
> +original project.")
> +    (home-page "https://github.com/yt-dlp/yt-dlp")
> +    (license license:public-domain)))
Otherwise LGTM, but haven't tested.

Regards,
Liliana

Liliana Marie Prikler <liliana.prikler@gmail.com> writes:

> Hi Robin,
>
> when pasting other stuff into a patch, one usually ought to do so after
> the "---" line before the summary and the first diff.

Ah, thanks for the reminder, it's been a minute since I used email for
patch submissions :) So IIUC, it's preferable to write:

> * gnu/packages/video.scm (yt-dlp): New variable. [...]
> ---
> 
> [RANDOM COMMENTS HERE]
> 
>  gnu/packages/video.scm | 61
> ++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 61 insertions(+)

rather than using "cut here" lines and the like.

> Am Mittwoch, den 13.10.2021, 08:46 -0400 schrieb Robin Templeton:
>> Liliana Marie Prikler <liliana.prikler@gmail.com> writes:
>> 
>> [...]
>> > > +              (snippet
>> > > +               '(begin
>> > > +                  ;; Delete the pre-generated files, except for
>> > > the man page
>> > > +                  ;; which requires 'pandoc' to build.
>> [...]
>> > Is this the same as for youtube-dl?  If not, we might want to give
>> > pandoc as native-input.  That will increase build times, but it
>> > ought not to increase closure size.
>> 
>> This is analogous to youtube-dl's pandoc avoidance; IMHO youtube-dl
>> ought to use pandoc as a native input, but I wanted to keep the
>> packaging as close to youtube-dl's as possible.
> Hence why I asked whether youtube-dl does the same.  Making that change
> would probably be out of scope for now.

+1

>> [...]
>> > > +                  (add-before 'build 'build-generated-files
>> > > +                    (lambda _
>> > > +                      ;; Avoid the make targets that require
>> > > pandoc.
>> > > +                      (invoke "make"
>> > > +                              "PYTHON=python"
>> > > +                              "yt-dlp"
>> > > +                              ;;"youtube-dl.1"   ; needs pandoc
>> > > +                              "completions")))
>> > > +                  (add-before 'install 'fix-the-data-directories
>> > > +                    (lambda* (#:key outputs #:allow-other-keys)
>> > > +                      (let ((prefix (assoc-ref outputs "out")))
>> > > +                        (substitute* "setup.py"
>> > > +                          (("'etc/")
>> > > +                           (string-append "'" prefix "/etc/"))
>> > > +                          (("'share/")
>> > > +                           (string-append "'" prefix
>> > > "/share/")))
>> > > +                        #t))))))
>> > Horizontal space is at a premium and you can save some.
>> 
>> I'm not sure where exactly this should use fewer columns, but I
>> squeezed the make invocation onto a single line.
> You fixed it by accident, given that you're now required to use
> substitute-keyword-arguments*, putting #:phases on an extra line as it
> ought to be.

Oh, I see what you meant now, thanks for clarifying.

>> v2 also adds three inputs needed for the program to run correctly
>> (updated based on efraim's yt-dlp package, <https://bpa.st/FJDA>;; in
>> that package, the extra libraries are propagated inputs, but adding
>> them as regular inputs appears to be sufficient).
> Can we somehow verify that this is indeed the case?  Normally we would
> wrap PYTHONPATH in such a case.

python-build-system wraps the yt-dlc script correctly, although (as
jgart explained to me on IRC) the Python library inputs would need to
become propagated inputs if users or future packages want to use yt-dlc
as a library.

Morgan's patch packages yt-dlc the way I'd package *youtube-dl* (using a
git origin, generating the manpage with pandoc, etc.) but yt-dlc doesn't
inherit from youtube-dl in their patch. The non-downloading-related part
of yt-dlc's test suite (~10% of tests) *does* run without failures,
unlike youtube-dl, so I enabled tests and added the pytest invocation
from Morgan's version; otherwise, no changes in v3 (besides using the
latest yt-dlc release), which I'll post as a follow-up.

I don't have a strong opinion about my patch vs. Morgan's for an initial
version; at some point I think youtube-dl should be packaged similarly
to Morgan's approach, which would allow the yt-dlc definition to be
simplified too, but my patch is written to inherit from youtube-dl as
it's packaged now. IMO the order of the changes doesn't matter much.

Thanks,
Robin