| Re: Nice ways to compile [message #43408] |
Mon, 11 April 2005 18:18  |
JD Smith
Messages: 850
Registered: December 1999
|
Senior Member |
|
|
On Mon, 11 Apr 2005 13:42:43 -0500, Michael A. Miller wrote:
> I make save files by creating an idl script that compiles all the
> codes that my application uses (or at least those that I can
> remember!). I include resolve_all and resolve_all, /class for
> all the classes that I need. Then I do a save and exit. So
> making a save set is just a command like "idl make_save_file.pro"
> (see below). This has all the elegance of makefile that is
> maintained by hand, which is to say, very little. I've
> considered trying to make a preprocessor that creates header
> files so I can use makedepend. If incremental compilation were
> possible (that is, loading compiled code, instead of having to
> compile in order to make code available), that would be useful,
> but with IDL, it doesn't seem neccessary.
>
> Usually I create my make_save_file.pro's from listings of all the
> *.pro files in the directories where I know I've got code
> components for a given application. That makes save files with
> cruft that is never used, but it hasn't (yet) left me with
> anything missing.
When creating a SAV file, I go through and compile all my code one
procedure at a time, as below. First I regularize the path, then I go
through all ".pro" files, pruning any rejects, and compile them. Only
then do I call RESOLVE_ALL, and write out the binary. Since path
caching tends to mess up dynamic path changes, I tend to do this in a
new IDL session.
Does anyone else get TRNLOG and SETLOG errors when
RESOLVE_ALL runs? As near as I can tell, these are some obsolete VMS
routines that seem to sneak in somehow (through NasaLIB, I believe).
JD
pro compile_cubism
FORWARD_FUNCTION TRNLOG
@cubism_dir
ps=path_sep()
bindir=filepath(ROOT=cubism_dir,'bin')
if file_test(filepath(ROOT=bindir,'cubism_vm.sav')) then $
file_delete,filepath(ROOT=bindir,'cubism_vm.sav')
;; Go one level up and compile everything
sourcepath=cubism_dir
sourcepath=strmid(sourcepath,0,strpos(sourcepath,ps,/REVERSE _SEARCH))
;; Normalize path
path=strsplit(!PATH,':',/EXTRACT)
wh=where((sp=strpos(path,ps+'nasa',/REVERSE_SEARCH)) ne -1)
nasa=strmid(path[wh[0]],0,sp[wh[0]]+5)
!PATH=expand_path('<IDL_DEFAULT>'+":+"+sourcepath+':+'+nasa)
files=file_search(sourcepath,'*.pro')
skip_files=['cubism_dir','cubism_version','compile_cubism', $
ps+'scraps'+ps,'CVS'+ps]
resolve_routine,'XManager',/COMPILE_FULL_FILE
skip=0
for i=0,n_elements(files)-1 do begin
for j=0,n_elements(skip_files)-1 do begin
if strpos(files[i],skip_files[j]) ne -1 then begin
print,'Skipping '+files[i]
skip=1
break
endif
endfor
if skip then begin
skip=0
continue
endif
print,'Compiling '+files[i]
routine=strmid(files[i],0,strpos(files[i],".pro",/REVERSE_SEARCH))
routine=strmid(routine,strpos(files[i],ps,/REVERSE_SEARCH)+1 )
resolve_routine,routine,/EITHER,/NO_RECOMPILE,/COMPILE_FULL_ FILE
endfor
resolve_all,/CONTINUE_ON_ERROR
save,/ROUTINES,FILENAME=filepath(ROOT=bindir,'cubism_vm.sav' )
end
|
|
|
|
| Re: Nice ways to compile [message #43410 is a reply to message #43408] |
Mon, 11 April 2005 11:42  |
mmiller3
Messages: 81
Registered: January 2002
|
Member |
|
|
I make save files by creating an idl script that compiles all the
codes that my application uses (or at least those that I can
remember!). I include resolve_all and resolve_all, /class for
all the classes that I need. Then I do a save and exit. So
making a save set is just a command like "idl make_save_file.pro"
(see below). This has all the elegance of makefile that is
maintained by hand, which is to say, very little. I've
considered trying to make a preprocessor that creates header
files so I can use makedepend. If incremental compilation were
possible (that is, loading compiled code, instead of having to
compile in order to make code available), that would be useful,
but with IDL, it doesn't seem neccessary.
Usually I create my make_save_file.pro's from listings of all the
*.pro files in the directories where I know I've got code
components for a given application. That makes save files with
cruft that is never used, but it hasn't (yet) left me with
anything missing.
Mike
;; make_save_file.pro
.compile first_file
.compile second_file
.compile third_file
;; ...more here as needed...
.compile last_file
.compile main
resolve_all
resolve_all, class=['med_imageobj', $
'registered_med_imageobj',$
'indypet_kinetics', $
'ipvis_kinetics', $
'showprogress' $
]
save, /routines, filename='main.sav'
exit
|
|
|
|
|
|
| Re: Nice ways to compile [message #43434 is a reply to message #43430] |
Sun, 10 April 2005 08:45 |
Ken Mankoff
Messages: 158
Registered: February 2000
|
Senior Member |
|
|
On Sun, 10 Apr 2005, Robert Barnett wrote:
> I'm looking for an easier way to compile my code into a .sav file.
Me too. I use the same process (roughly). The method seems to work,
but it also seems fragile.
> + Complile the package and depenancies to a save file and run in
> an IDL VM "sandbox" environment.
> I expect to be able to compile my IDL packages with a single UNIX
> script (step marked as a +). I also expect that my compilation
> depends solely on the contents of the filesystem rather than the
> current state of the IDL command prompt. This ensures that I can
> always return to a snapshot of my idl directory as it was when I
> compiled my code.
I think it is tricky to get it to rely on the contents of the
filesystem. There are known issues when it will not track down
dependencies that you or I would see reading the code (is this what
you mean by filesystem?). Hence your request for a compiler
directive.
> For example, I currently compile like the following:
I use IDL with other software that I build from emacs/CLI by calling
make on a standard *nix Makefile. So 1 command: "make". At some
point in the dependency chain I call
idl make.pro
which is
; makefile for EVA
; produces eva.sav
.reset_session
.com my_map_set
.com eva
.com eva_earth__define
.com eva_ocean__define
.com eva_data__define
.com eva_image__define
;;; this part grows as my code library grows.
RESOLVE_ALL
spawn, 'echo $HOSTNAME', h
if h = 'random' then save, /routines, EMBEDDED=..., file='eva.sav' else $
save, /routines, file='eva.sav' ; no license
$cp eva.sav ../IDL/bin/
exit
; is there any way to determine if I am a real interactive IDL
; session or a unix CLI "$ idl foo.pro" version? Because
; this exit should be conditional :)
> Perhaps I am looking at this problem completely wrong. I would
> expect there to be some way I can append a compiler directive
> which indicates that blah__define.pro requires
> 'idlgrlegend__define'. Compiler directives are not really in the
> IDL idiom, so I'm wondering what other choices I might have.
They are in the idiom. See ?COMPILE_OPT
But none does what you want that I know of.
I'm thinking there is some way (with a shell script and grep?) to
solve your dependencies automagically, until IDL does it correctly
itself.
-k.
--
http://spacebit.dyndns.org/
|
|
|
|
|
|
| Re: Nice ways to compile [message #43526 is a reply to message #43494] |
Sat, 16 April 2005 02:35 |
biophys
Messages: 68
Registered: July 2004
|
Member |
|
|
I use skip_routines and it works fine:
resolve_all, class=['MYVMCONSOLE', 'FSC_PSCONFIG'], $
skip_routines=['TRNLOG','DELLOG','SETLOG']
Jeff Guerber wrote:
> On Mon, 11 Apr 2005, JD Smith wrote:
>
>> Does anyone else get TRNLOG and SETLOG errors when
>> RESOLVE_ALL runs? As near as I can tell, these are some obsolete
VMS
>> routines that seem to sneak in somehow (through NasaLIB, I
believe).
>
> Yes, also DELLOG. I forget exactly where they were, but there
seemed
> to still be references to them within IDL itself. I created small
dummy
> procedures:
>
> pro setlog
> end
>
> etc., just so they'd have definitions.
>
> (I can't really add anything on the real issue. We have a simple
> script that makes a list of all the .pro files in the directory and
writes
> an IDL procedure with a .compile for each, which is a solution
someone
> else has already mentioned.)
>
> Jeff Guerber
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
