rpmlint says the following (certainly) relevant comments:

ncl.i386: E: executable-sourced-script /etc/profile.d/ncarg.csh 0755
ncl.i386: W: devel-file-in-non-devel-package /usr/lib/ncarg/ncarg/graphcaps/aed.a
ncl.i386: E: executable-sourced-script /etc/profile.d/ncarg.sh 0755
ncl.i386: W: non-standard-group Engineering/Scientific
ncl.src: W: non-standard-group Engineering/Scientific
ncl.src: W: strange-permission ncarg.csh 0755
ncl.src: W: strange-permission ncarg.sh 0755

Group could be Applications/Engineering.

Thanks, should be fixed now.

* Wed Nov 14 2007 - Orion Poplawski <orion.com> - 5.0.0-2
- Fixup profile.d script permissions, Group, move aed.a to devel

http://www.cora.nwra.com/~orion/fedora/ncl-5.0.0-2.fc8.src.rpm
http://www.cora.nwra.com/~orion/fedora/ncl.spec

The remaining rpmlint complaints are:
  ncl.x86_64: W: non-conffile-in-etc /etc/profile.d/ncarg.csh
  ncl.x86_64: W: non-conffile-in-etc /etc/profile.d/ncarg.sh
which are both OK as I understand things.

I know this isn't going to be a pretty package, what with needing both imake and
fortran to build, but the end result doesn't seem terribly bad.

Some of the binaries seem destined to conflict with something (fontc, med,
psplit, etc.) but this package has been around for so long that it would be
pointless to think of renaming any of them.

Now, on to the stuff in %_libdir:

First, why is it all in %_libdir/ncarg/ncarg?  Why not just one ncarg?

What are the .o files in %_libdir/ncarg/ncarg/robj for?

Your -devel package should provide ncl-static = %{version}-%{release}.

But really, for a crufty piece of software like this, this package isn't bad.

There are some items I don't understand in Site.local. Why is there
#define BuildUdunits FALSE
#define BuildV5D TRUE
#define BuildNetCDF4 TRUE

Also, I haven't verified, but in the README it is said that the 
CcOptions,... can be redefined in Site.local.

Another thing is that it seems to me that it would be better
to set NCARG_ROOT and NCARG_LIB to the right default in 
config/Project or config/Site.local instead of using 
environment variables.

I can do patches to do what I propose.

Also looks like imake is not used, instead they use ymake,
though this may be the same?

(In reply to comment #3)
> Now, on to the stuff in %_libdir:
> 
> First, why is it all in %_libdir/ncarg/ncarg?  Why not just one ncarg?

We put the .a in %_libdir/ncarg to avoid (possible?) conflicts.  Once that has
been done, you end up with ncarg/ncarg, and it's much too much of a mess to try
to change.

> What are the .o files in %_libdir/ncarg/ncarg/robj for?

Looks like they are various optional features that can be turned on through
flags in ncargfcc, etc. when compiling projs.  I've moved the -devel package.

> Your -devel package should provide ncl-static = %{version}-%{release}.

Correct, done.

> But really, for a crufty piece of software like this, this package isn't bad.

Thanks :-)

What about my comments?

Also I have something really strange, there is no ncl executable in
the rpm I rebuilded??? Do you see the same? Did I messed something?
I will investigate in any case.

(In reply to comment #4)
> There are some items I don't understand in Site.local. Why is there
> #define BuildUdunits FALSE

Changed.

> #define BuildV5D TRUE

Changed.

> #define BuildNetCDF4 TRUE

Changed.

> Also, I haven't verified, but in the README it is said that the 
> CcOptions,... can be redefined in Site.local.

Yeah, but you end up with a lot of warnings about redefining macros during the
dependency generation stage.  I don't like lots of spurious warnings.
  
> Another thing is that it seems to me that it would be better
> to set NCARG_ROOT and NCARG_LIB to the right default in 
> config/Project or config/Site.local instead of using 
> environment variables.

I found it impossible to do because of how the install system works.

Currently testing some patches, should have more soon...

(In reply to comment #7)
> Also I have something really strange, there is no ncl executable in
> the rpm I rebuilded??? Do you see the same? Did I messed something?
> I will investigate in any case.

I have certainly found the error:

gcc -ansi -fPIC -O2 -g -pipe -Wall -Wp,-D_FORTIFY_SOURCE=2 -fexceptions
-fstack-protector --param=ssp-buffer-size=4 -m32 -march=i386 -mtune=generic
-fasynchronous-unwind-tables -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -O 
-I../../.././include -I/usr/include/netcdf-3  -DLinux -DMAJOR=2 -DUSE_NETCDF4
-DBuildHDFEOS -DBuildGRIB2    -D_POSIX_SOURCE -D_XOPEN_SOURCE -DByteSwapped
-DNeedFuncProto    -c -o scanner.o scanner.c
ncl.l: In function '_NclParseString':
ncl.l:76: warning: implicit declaration of function 'yyparse'
ncl.l: In function '_NclResetScanner':
ncl.l:199: warning: implicit declaration of function 'isatty'
ncl.l: In function 'yylex':
ncl.l:273: warning: implicit declaration of function '_NclCallPromptFunc'
ncl.l:805: warning: ignoring return value of 'fwrite', declared with attribute
warn_unused_result
lex.yy.c: In function 'yy_get_next_buffer':
lex.yy.c:2099: error: 'yy_current_buffer' undeclared (first use in this function)
lex.yy.c:2099: error: (Each undeclared identifier is reported only once
lex.yy.c:2099: error: for each function it appears in.)
make[4]: *** [scanner.o] Error 1

and make happily continues...

Definitely what I hate about the ncarg/ncl build system.  I've fixed that error
and am on to another.  Will post when I have a working version.

This should actually build ncl:

http://www.cora.nwra.com/~orion/fedora/ncl-5.0.0-4.fc8.src.rpm
http://www.cora.nwra.com/~orion/fedora/ncl.spec

It also uses ATLAS instead of the packaged blas and lapack libraries but hasn't
been tested.  I'm getting some undefined symbol errors that I need to look into too.

This builds wrapit77:

http://www.cora.nwra.com/~orion/fedora/ncl-5.0.0-5.fc8.src.rpm
http://www.cora.nwra.com/~orion/fedora/ncl.spec

I think that's it for the major build issues.

I made some work on ncl to correct some issues I see. noarch
stuff isn't in datadir, paths could be fixed. I ran into
other issues while doing that. Here are the patches.

Created attachment 293771 [details]
use datadir for noarch files and other fixes

Created attachment 293772 [details]
corresponding patch for Site.local, also use more rpm macros

Created attachment 293773 [details]
build file patch to install noarch files in NCARG_ROOT/share

I think that this patch could go upstream.

Created attachment 293774 [details]
don't have install depend on target

I am not sure that it is right, but otherwise the in source libraries
have ranlib rerun on them just before they are installed, causing
the binaries that depend on these libraries to be 
relinked before installation.

Created attachment 293776 [details]
specific build target are put after generic ones

This allows to have the 'all' target be the first one,
instead of one of the scripts target, such that the
default is to remake all the scripts.

Created attachment 293777 [details]
link against netcdff for fortran linking

Patrice -

  Thanks for all your hard work here!  I've applied the patches.

http://www.cora.nwra.com/~orion/fedora/ncl-5.0.0-8.fc8.src.rpm
http://www.cora.nwra.com/~orion/fedora/ncl.spec

* Wed Feb  6 2008 - Orion Poplawski <orion.com> - 5.0.0-8
- Move examples into separate sub-package

* Fri Feb  1 2008 - Patrice Dumas <pertusus> - 5.0.0-7
- put noarch files in datadir
- avoid compilation in %%install

* Mon Jan 14 2008 - Orion Poplawski <orion.com> - 5.0.0-6
- Make BR hdf-devel >= 4.2r2.

A remaining issue:

I cannot use the source url, it leads to unauthorized.

Minor issues:

The Version isn't in sync with %changelog.

The buildrequires imake isn't needed (unless I am wrong).

Missing dots at the end of devel and examples %descriptions.

I suggest calling Site.local along ncl-Site.local instead.

I also suggest shortening a bit the devel package summary. 

You must register for an account at http://www.earthsystemgrid.org/ before
downloading the source.  I complained about this on the NCL list, but they feel
that their government contracts dictate that they try to collect usage
statistics and they feel this is the only option available for them.  Made a
note of this in the spec.

Upped the release number.

imake provides makedepend which is used.

added dots

Fixed up devel summary/description

Not sure I understand about Site.local.  Seems fine the way it is.

(In reply to comment #22)
> You must register for an account at http://www.earthsystemgrid.org/ before
> downloading the source.  I complained about this on the NCL list, but they feel
> that their government contracts dictate that they try to collect usage
> statistics and they feel this is the only option available for them.  Made a
> note of this in the spec.

This is really, really (really) annoying. Moreover it doesn't make 
sense since anybody can repost the archive and deployments are
better done through packages. I personnally don't want to approve a 
package with such a restriction done upstream.

Otherwise I think that the package is acceptable.

> imake provides makedepend which is used.

Not a blocker, but a comment would be nice.
 
> Not sure I understand about Site.local.  Seems fine the way it is.

My point is that Site.local is not really specific of ncl, one
could imagine that a file with the same name is in another package.
Disambiguating the same in SOURCES may be useful in settings where
the same SOURCES directory is used for all packages.

As I said, only a suggestion.

(In reply to comment #23)
> (In reply to comment #22)
> 
> > imake provides makedepend which is used.
> 
> Not a blocker, but a comment would be nice.

Done.
  
> > Not sure I understand about Site.local.  Seems fine the way it is.
> 
> My point is that Site.local is not really specific of ncl, one
> could imagine that a file with the same name is in another package.
> Disambiguating the same in SOURCES may be useful in settings where
> the same SOURCES directory is used for all packages.
> 
> As I said, only a suggestion.

Ah, I always forget about shared SOURCES directories.  Changed.

http://www.cora.nwra.com/~orion/fedora/ncl-5.0.0-9.fc8.src.rpm
http://www.cora.nwra.com/~orion/fedora/ncl.spec

* Mon Feb 18 2008 - Orion Poplawski <orion.com> - 5.0.0-9
- Rename Site.local to Site.local.ncl
- Add comment for imake BR

Thanks again Patrice for your work.

Anyone else out there willing to approve a package as it is (with having to
register to download source)?

After reviewing all the comments here it seems that Patrice has done a
very thorough review.  The only thing missing seems to be the "source 
matches upstream" check which was omitted due to Pat's (understandable) 
stance on the annoying must-register-to-download primary web site.  After
registering I can confirm: 

 + source matches upstream with sha1sum:
    eca618a1179356e2950608da0e9ca210f1caa1a8  ncl_ncarg_src-5.0.0.tar.gz
    eca618a1179356e2950608da0e9ca210f1caa1a8  ../ncl_ncarg_src-5.0.0.tar.gz
 + specfile is clean and legible
 + license appears to be correct
 + builds in mock on F8 x86_64

Since there appear to be no blockers and since Patrice has already done a 
very thorough review, I'd like to APPROVE this package.

New Package CVS Request
=======================
Package Name: ncl
Short Description: NCAR Command Language and NCAR Graphics
Owners: orion
Branches: F-8 F-7 EL-5
InitialCC: 

cvs done.

This is built, closing.