[petsc-dev] a new script "install" ? Re: SLEPc headers and integration into PETSc

Barry Smith bsmith at mcs.anl.gov
Tue Jun 21 15:11:41 CDT 2011


On Jun 21, 2011, at 2:02 PM, Andrew Knyazev wrote:

> I do not propose to remove the possibility for the user to type
> "configure" and "make" separately, if they really want to. Just add a
> new possibility to do everything at once, like what you apparently
> describe by TOPSInstaller. You should keep a non-gui version of it, too,
> for remote installs.

   That is what a browser is for :-).

   Barry

> 
> On 06/21/2011 01:50 PM, Barry Smith wrote:
>> 
>> Run ./bin/TOPSInstaller.py     -- the next generation of this should be a javascript  gui beasty that runs in your browser.
>> 
>> 
>> Regarding the ./configure; make  model. If we didn't have this all the gnu/linux bigots would give us a hard time (which they do already). In other words, it is what most people expect. 
>> 
>>   Barry
>> 
>> On Jun 21, 2011, at 1:46 PM, Andrew Knyazev wrote:
>> 
>>> On 06/21/2011 01:29 PM, Barry Smith wrote:
>>>> 
>>>> On Jun 21, 2011, at 1:07 PM, Andrew Knyazev wrote:
>>>> 
>>>>> On 06/21/2011 12:24 PM, Jose E. Roman wrote:
>>>>>>> SLEPc has the headers not in the same directory as PETSc? Why?
>>>>>> SLEPc could be installed by a user that does not have write permission in PETSc's directory.
>>>>> 
>>>>> PETSc gives so many different configure options, so it is quite rare to
>>>>> see it installed by a sysadmin nowadays, in my experience.
>>>>> 
>>>>> At any rate, it would be nice to have a tighter integration of SLEPc
>>>>> into PETsc, ideally, simply to be able to install SLEPSc by using
>>>>> --download-slepsc=1 option in PETSc. I am sure that it would
>>>>> significanly increase the number of SLPEc users.
>>>> 
>>>> 
>>>> To do this we need to broaden our Packages concept to have both pre and post packages. Currently we handle pre packages (that PETSc uses) pretty well but do not handle post packages (that use PETSc). Prometheus is this weird thing that is partially pre and partially post and is handled a bit too ad hocly.
>>>> 
>>>> In some ways post packages are pretty easy, we just need to set up the infrastructure. 
>>>> 
>>>> Since the user doesn't care about pre and post we'd want to support the same --download-xxx syntax in both cases (with some way of passing optional arguments) and, of course, as Jed points out additional -download-xxx can be used after a build.
>>>> 
>>>>  Barry
>>> 
>>> 
>>> For the user, could you perhaps add a new script "install" which would
>>> just do everything: configure, make PETSc, install PETSc, and all
>>> necessary pre and post packages, plus compile all examples? I could
>>> never understand why PETSc requires the user to type "make" separately
>>> and then also compile every individual example.
>>> 
>>> Of course, typing "make" gives the user a false feeling that they know
>>> and control what they are doing. But this feeling goes away quickly, and
>>> the moral value of these few high moments is not that great anyway.
>> 
> 




More information about the petsc-dev mailing list