test_savetxt_qp fails with ifort

[pdebuyl]

Using "ifort (IFORT) 18.0.1 20171018" on linux x86_64, test_savetxt_qp fails

The routine qsavetxt writes one number per line, which is allowed by the star format.

I propose a fix in https://github.com/pdebuyl/stdlib/tree/qsavetxt_format_string , tested with said ifort and with gfortran 6.4.0 and 8.3.0.

I have not opened a PR yet, as I don't know if ifort compatibility is considered necessary at this point. I would support keeping stdlib working with several compilers early on.

[jvdp1]

Is this problem related to #81? Maybe you could implement the same strategy as in #81 to avoid qp with Intel compilers and open a new PR.

However, I think your fix could be also useful for very large arrays (to avoid multiple lines for one row of an array; I didn't test it yet). However, this implies a fixed format (e.g., 'f40.34), and the same should be done for spanddp. Another option would be to discuss the API of loadtxtandsavetxt` to allow the user to specify its own format. See #16 ( comment ) and #14 for some discussions.

[pdebuyl]

It is orthogonal to #81 I believe. ifort supports a qp type, it is only the write operation that does not output a "ncol x nrow" format.

Issue #16 is of course related but consider extensions of loadtxt and savetxt. Here, I just fix the currrent version :-)

I write a format string to avoid relying on a fixed format. The current format string can hold up to 999999 for the number of columns. At some point users should use HDF5 or another solution anyway.

[certik]

Can you please submit a PR? Compatibility with ifort is essential.

…

On Mon, Jan 6, 2020, at 2:39 AM, Pierre de Buyl wrote: Using "ifort (IFORT) 18.0.1 20171018" on linux x86_64, test_savetxt_qp fails The routine qsavetxt writes one number per line, which is allowed by the star format. I propose a fix in https://github.com/pdebuyl/stdlib/tree/qsavetxt_format_string , tested with said ifort and with gfortran 6.4.0 and 8.3.0. I have not opened a PR yet, as I don't know if ifort compatibility is considered necessary at this point. I would support keeping stdlib working with several compilers early on. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#88?email_source=notifications&email_token=AAAFAWEVRGZIG544BQMK5N3Q4L33RA5CNFSM4KDB4AAKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IEFVYTA>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAFAWFIF66C7CBXVP6IKLLQ4L33RANCNFSM4KDB4AAA>.

[scivision]

This would be addressed by #93

[pdebuyl]

@scivision ifort supports qp, it is only the * formatting that fails to generate colummar data files.

[pdebuyl]

It is fixed actually, so I close the issue.

[certik]

This was fixed by #89.