Bug#968531: littler: probably shouldn't use install -s

2020-08-16 Thread Dirk Eddelbuettel


Hi Simon,

On 16 August 2020 at 23:53, Simon McVittie wrote:
| Source: littler
| Version: 0.3.11-1
| Severity: minor
| 
| A bug report against fteqcc (#968524) prompted me to look for other
| instances of the same anti-pattern on codesearch.debian.net.
| 
| d/rules in littler invokes 'install -s'. However, 'install -s' is usually
| only harmful for a Debian package, not helpful: it discards debugging
| symbols that could otherwise have gone into a -dbgsym package, when
| cross-compiling it uses the wrong strip(1) implementation (from the build
| rather than host architecture), and it results in the nostrip option in
| DEB_BUILD_OPTIONS being ignored (see Policy §4.9.1). Unconditionally
| using INSTALL_PROGRAM = install would likely be better.

Thanks for the heads-up. Easy enough to fix in

  common-binary-post-install-arch::
  dh_installdirs usr/bin usr/share/man/man1
  install -s -v -m 0644 $(debRlib)/littler/bin/r
$(CURDIR)/debian/$(package)/usr/bin/
  install-v -m 0644 $(debRlib)/littler/man-page/r.1 
$(CURDIR)/debian/$(package)/usr/share/man/man1/

The package is a bit of an odd-ball -- I am the upstream author, and we at
first had "non-standard" distribution (in that it is an R package but wasn't
going to CRAN).

There is an added difficulty in that R packages only ever install in their
own director and below it, but this one is meant for common $PATH -- which is
where the line above comes from as R has no means to install in, say,
/usr/bin or /usr/local/bin

I made the change in my package sources, this will go out with the next littler 
update.

Best, Dirk

-- 
https://dirk.eddelbuettel.com | @eddelbuettel | e...@debian.org



Bug#968531: littler: probably shouldn't use install -s

2020-08-16 Thread Simon McVittie
Source: littler
Version: 0.3.11-1
Severity: minor

A bug report against fteqcc (#968524) prompted me to look for other
instances of the same anti-pattern on codesearch.debian.net.

d/rules in littler invokes 'install -s'. However, 'install -s' is usually
only harmful for a Debian package, not helpful: it discards debugging
symbols that could otherwise have gone into a -dbgsym package, when
cross-compiling it uses the wrong strip(1) implementation (from the build
rather than host architecture), and it results in the nostrip option in
DEB_BUILD_OPTIONS being ignored (see Policy §4.9.1). Unconditionally
using INSTALL_PROGRAM = install would likely be better.

smcv