On 8 May 2019, at 14:08, Patrick Sanan 
<[email protected]<mailto:[email protected]>> wrote:



Am Mi., 8. Mai 2019 um 14:03 Uhr schrieb Matthew Knepley via petsc-dev 
<[email protected]<mailto:[email protected]>>:
On Wed, May 8, 2019 at 7:41 AM Hapla Vaclav via petsc-dev 
<[email protected]<mailto:[email protected]>> wrote:
Hello

I just encountered their manpages and it's a mess. I think in case of file I/O, 
a user should be deliberate about filename and other settings. Sometimes less 
is more and I think this is the case. Why anybody should use 
PETSC_VIEWER_BINARY_(comm) and then set the filename with environment variable? 
Better to use PetscObjectViewFromOptions for instance (PETSC_OPTIONS can be 
used if one wants to use env var).

(And BTW in case of HDF5, the option -viewer_hdf5_filename is not processed 
anywhere.)

Any objections about removing these?

I am fine with it. The new system is better.

Since these are user interface, we should deprecate them somehow. Will the 
normal deprecation work?
These are functions (albeit weirdly-named ones which don't return error codes), 
so one would assume that the usual function deprecation procedure would work, 
and then things like PETSC_VIEWER_BINARY_WORLD would also show up as deprecated.

... Which I think is correct. I meant also PETSC_VIEWER_BINARY_{SELF,WORLD}.

I would even suggest to keep just PETSC_VIEWER_STDOUT* stuff (used most often 
and with least additional settings needed), or nothing at all.

Users should be encouraged to use stuff controllable from the options database.
I think changing the PetscViewerType by rewriting the hard-wired PETSC_VIEWER_* 
call sounds legacy.
If one insists on setting up everything using API, still it is more flexible to 
just use PetscViewerSetType and/or PetscViewerSetFromOptions.

Thanks,
Vaclav


  Thanks,

    Matt

Vaclav


--
What most experimenters take for granted before they begin their experiments is 
infinitely more interesting than any results to which their experiments lead.
-- Norbert Wiener

https://www.cse.buffalo.edu/~knepley/<http://www.cse.buffalo.edu/~knepley/>

Reply via email to