Bug#999785: Bogus Lintian warning built-using-field-on-arch-all-package affects prospective package firmware-carl9170

[Julian Gilbey <jdg@debian.org>]

retitle 999785 built-using-field-on-arch-all-package emitted for some Arch: all packages that require it
thanks

On Mon, Jul 10, 2023 at 06:01:40AM +0000, John Scott wrote:
> Lintian asserts that having Built-Using on an Arch: all package is always incorrect. Debian Policy permits and often requires having Built-Using on an Arch: all package. This is the situation with carl9170fw, a GPL-2.0-only-licensed binary that bakes in several static libraries that need to have their sources kept around. This is a firmware package, however, so it's Arch: all despite being written in C and producing a bare-metal binary.
> [...]
> 
> Furthermore, the far more likely problem with Built-Using is that a package forgets to use it when it should. So when a package *does* remember to set the Built-Using field, it's typically the result of long consideration and the maintainer should be given the benefit of the doubt.

Fully agree; as noted in #1029633, sphinxdoc requires this, despite
generating Arch: all packages.  The tag description:

N:   The stanza for an installation package in debian/control declares a
N:   Built-Using field even though the package is declared as Architecture:
N:   all. That is incorrect.
N:   
N:   The Built-Using field is only used architecture-specific packages. Please
N:   remove the Built-Using field from the indicated location.

is simply wrong.

Conversely, it would be helpful to have a check for packages using
sphinxdoc (either B-D on dh-sequence-sphinxdoc or using dh --with
sphinxdoc) that they *are* specifying Built-Using:
${sphinxdoc:Built-Using}.

Best wishes,

   Julian