improve error message for find region

[amanzi]

improve error message for find region

[lipnikov]

Since it is ATS and not Amanzi, more details would be helpful. For example, this exception could be handled inside ATS's main().

[ecoon]

It's not an ATS thing though since any physics can call get_set_*() calls. I thought it was handled in the geometric model, but apparently not. I can deal with it, but any opposition to putting it there?

[lipnikov]

valid_set_name() does the job you need. Of coarse, we can check name validity every time we call get_set_*() and replace one exception with another. But this will be the fourth level of control in Amanzi on the top of (1) Akuna that creates an input file using existing regions, (2) converter that performs many checks and (3) InputAnalysis that verifies mesh sets.

[ecoon]

No, this is clearly an ask forgiveness not permission. "try" has no overhead on almost all compilers, only costs more if exceptions are thrown, at which point the code is dying anyway so who cares about a bit of overhead.

[lipnikov]

Where should we draw a line? Check validity regions of input parameters (like RCP pointers) in all subroutines and throw an exception when checks fail? I vote for using valid_set_name() and improving error messages in this subroutine. But this is Rao's code and I'll follow his advice.

[raovgarimella]

If you use valid_set_name, then you will get a more meaningful error message. It is obviously a non-zero expense so perhaps one can enclose it within a #ifdef DEBUG statement or call AMANZI_ASSERT on its result?

[ecoon]

Alternately we can try, catch the exception, then call valid_set_name().
Cost is free when the set is valid, and we still get the more useful error
message.

In terms of where to draw the line, I believe that error messages should be
handled in the library/component which has the error. Validating the input
spec is great, but not robust, and not self-contained.

For instance, if the mesh class gets pulled out as a library, calling it
maybe Jolly or something, then the input spec parser is worthless and the
error message is again useless. Components should be as self contained as
possible.

Furthermore, your example isn't a good one -- a bad Ptr/RCP is a developer
bug. A bad set name is a user bug. We should be much more careful about
error messages with regards to the latter.

Ethan

On Fri, Oct 21, 2016 at 1:43 PM Rao Garimella [email protected]
wrote:

If you use valid_set_name, then you will get a more meaningful error
message. It is obviously a non-zero expense so perhaps one can enclose it
within a #ifdef DEBUG statement or call AMANZI_ASSERT on its result?

—
You are receiving this because you were assigned.
Reply to this email directly, view it on GitHub
#47 (comment), or mute
the thread
https://github.com/notifications/unsubscribe-auth/AE43-2w0GABCLURKYYVv-NiRUjoHX08Rks5q2RXSgaJpZM4Kcf0_
.

[ecoon]

Thanks for doing this Konstantin, was on my todo list and just never got important enough.

…

On Wed, Jun 28, 2017 at 1:30 PM, lipnikov ***@***.***> wrote: Closed #47 <#47> via 3bedc7f <3bedc7f> . — You are receiving this because you were assigned. Reply to this email directly, view it on GitHub <#47 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AE43-6G8BwfsBDvt0sGwesAAqNq9AGNwks5sIo22gaJpZM4Kcf0_> .

--

------------------------------------------------------------------- Ethan Coon http://www.lanl.gov/expertise/profiles/view/ethan-coon -------------------------------------------------------------------