installation of info files

[Till Maas]

Hiyas,

the guidelines do not mention how to install .info files the right way. At the 
moment there seem to be several scriptlets used to install them in %post 
and %preun. I want to suggest to define a rpm-macro that handles info files 
in the right way which also would reduce redundant lines of code in 
spec-files. Something like "%info" in the files section would be most useful 
in my oppionion, like it already exists for documentation (%doc).

What are your oppionions about this?

Regards,
Till

[Ralf Corsepius]

On Wed, 2006-07-12 at 22:28 +0200, Till Maas wrote:
>  Hiyas,
>  
>  the guidelines do not mention how to install .info files the right way. At the 
>  moment there seem to be several scriptlets used to install them in %post 
>  and %preun. I want to suggest to define a rpm-macro that handles info files 
>  in the right way which also would reduce redundant lines of code in 
>  spec-files. Something like "%info" in the files section would be most useful 
>  in my oppionion, like it already exists for documentation (%doc).
>  
>  What are your oppionions about this? 
Introducing such macros is a double sided sword:
* On one hand, they to some extend ease writing specs.

* On the other hand:
- Once such macros have been introduced, it's very difficult to get rid
of them, should one want to abandon such macros. In many cases, you will
have to continue to provide them for a very long (many years)
transitional time period. Rpm itself is full of semi-functional/once
thought to be useful macros, nowadays almost nobody uses.

- Vendor-specific macros are a major obstacle to spec file portability.
This isn't of much importance within a distribution, but is of
importance for 3rd parties who want to provide add-on/drop-in rpms.
SuSE and Mandrake users probably are aware about the (often
unresolvable) problems such macros introduce, because these distros tend
use them extensively.

=> IMO, in general such macros should be avoided as far as possible.
In an ideal world all macros should either be provided by vendor
independent packages, or as part of by rpm itself.

Ralf

[Ville Skyttä]

On Thu, 2006-07-13 at 12:51 +0200, Ralf Corsepius wrote:

>  Introducing such macros is a double sided sword: 
[...]

Agreed.

Regarding the case at hand, in case someone's thinking about submitting
an RFE somewhere, FWIW Mandriva has had macros for installing and
removing info files for a long time, they're defined as:

%__install_info         /sbin/install-info

%_extension             .bz2

%_install_info()        if [[ -f %{_infodir}/%{1}%{_extension} ]];then %{__install_info}
%{_infodir}/%{1}%{_extension} --dir=%{_infodir}/dir;fi \
%{nil}

%_remove_install_info() if [ "$1" = "0" ]; then if [[ -f
%{_infodir}/%{1}%{_extension} ]];then %{__install_info} %{_infodir}/%{1}%{_extension}
--dir=%{_infodir}/dir --remove ;fi; fi \
%{nil}

The naming (%_extension, %_remove_install_info) is unfortunate and other
nitpicks could be said, but perhaps these are useful as a starting
point.

[Enrico Scholz]

ville.skytta(a)iki.fi (Ville Skyttä) writes:

>  Regarding the case at hand, in case someone's thinking about
> submitting
>  an RFE somewhere, FWIW Mandriva has had macros for installing and
>  removing info files for a long time, they're defined as:
>
>  %__install_info         /sbin/install-info
>
>  %_extension             .bz2 
install-info is smart enough to add at least '.gz' when needed. Hence
the %_extension macro is not needed.


>  %_install_info()        if [[ -f %{_infodir}/%{1}%{_extension}
> ]];then %{__install_info} %{_infodir}/%{1}%{_extension} --dir=%{_infodir}/dir;fi \
>  %{nil}
>
>  %_remove_install_info() if [ "$1" = "0" ]; then if [[ -f
> %{_infodir}/%{1}%{_extension} ]];then %{__install_info} %{_infodir}/%{1}%{_extension}
> --dir=%{_infodir}/dir --remove ;fi; fi \
>  %{nil} 

I used

| %__install_info		/sbin/install-info --info-dir=%_infodir
| %__install_info_post          g() { %__install_info "$@" || :; }; g
| %__install_info_preun         g() { %__install_info --delete "$@" || :; };
test "$1" != 0 || g

some time ago. I do not see much sense in the test-for-existence check;
when a packager uses

| %__install_info_post %_infodir/zsh.info

he knows that 'zsh.info' exists there (else, package would be broken).


I am not sure about automatic detection of pre/postun requires; perhaps
a similar approach like for fedora-usermgmt can be chosen and

| %FE_INSTALL_INFO_REQ	\
| Requires(pre):	%__install_info	\
| Requires(postun):	%__install_info

defined which is applied like

| BuildRequires: package-with-these-macros
| %{?FE_INSTALL_INFO_REQ}




Enrico

[Christian Iseli]

enrico.scholz(a)informatik.tu-chemnitz.de said:
>  I do not see much sense in the test-for-existence check;
>  when a packager uses
>  | %__install_info_post %_infodir/zsh.info
>  he knows that 'zsh.info' exists there (else, package would be broken).  
Hrm, not sure here, but what about --excludedocs ?

					Christian

[Enrico Scholz]

Christian.Iseli(a)licr.org writes:

> > I do not see much sense in the test-for-existence check;
> > when a packager uses
> > | %__install_info_post          g() { %__install_info "$@" || :; }; g
> > | ...
> > | %__install_info_post %_infodir/zsh.info
> > he knows that 'zsh.info' exists there (else, package would be broken). 
>
>  Hrm, not sure here, but what about --excludedocs ? 
Not really a problem; the command has a '|| :' making the scriptlet
succeed. When you would want to handle missing files (--excludedocs or
--excludepaths), you would have to handle read-only %netsharedpath
mounts too (which would make install-info fail).


My (real) original macro was

| %define __isDirectoryShared	f() { rpm --eval '%%{_netsharedpath}' | awk -v
y="$1" '{ split($1, x, ":"); for (i in x) if (index(y,x[i])==1)
exit 0; exit 1; }'; }; fi
| %define __install_info	( %__isDirectoryShared %_infodir ) || /sbin/install-info
--info-dir=%_infodir

but it is not really worth the effort because e.g. 'rpm' might be
missing in chroot environments and rpm maintainer refuses to add
functionality allowing to detect whether path is covered by
%netsharedpath resp. missing due to --exclude*.

Perhaps a '2>/dev/null' should be added.



Enrico