Discard trailing entries of a line in a MM-format file #1628
Conversation
ginkgo-bot
added
reg:testing
With the current |
There was a problem hiding this comment.
nit: could you also remove the nnz in the error message in line L577-L579 of core/base/mtx_io.cpp
It is for the dense, so there's no need for nnz.
could you also share the real case (I assume from some application) which does not fit the expected MM-file?
I am thinking whether we should ignore the additional entry or just throw the error unless they use some % as the comment?
Could you update the documentation with this behavior?
|
The "real case" was just me making a mistake in the dense/"array" format for Matrix Market (vs coordinate). I didn't check that the entries were supposed to be in column-major order, and instead provided them row-major (the way you would print out the matrix to look at it visually) and without the correct one-per-line formatting. Because I still had the correct total number of entries, Ginkgo did not throw an error (whitespace between values was treated the same as newlines), so I didn't realize the values were being interpreted incorrectly and therefore causing problems with my test. So it was a case of user error for sure, but the idea behind this PR was that Ginkgo could help users by making this a failure mode -- if you do not have the correct number of lines, it will be an error, and you will realize your file formatting is incorrect. It has the benefit of more strictly adhering to the MM standard, apart from cases of user error. |
- formatting - fix error message Co-authored-by: Thomas Grützmacher <thomas.gruetzmacher@tum.de> Co-authored-by: Yu-Hsiang M. Tsai <yhmtsai@gmail.com>
MarcelKoch
force-pushed
the
mm-io-discard-rest-of-row
branch
from
9a60b3b
to
c28124a
Compare
I don't think there is anything to update. This just fixes the behavior for our previous wrong parsing. |
MarcelKoch
added
1:ST:ready-to-merge
This PR will discard any character in a line of a MM-format file after the values have been read in. This addresses #1627, using multiple columns within the MM-file will now lead to an exception. However, the exception is thrown by the
read_entryfunction and thus the exception message is not very helpful.