Franck HOUSSEN

@jgmbenoit: `howmny` seems to be `character*1` https://github.com/opencollab/arpack-ng/blob/872df83b9e4f493f4bf68457a96edbcafafd5e4c/SRC/dseupd.f#L54 Did you try that? ``` - char *all="All"; + char *all="A"; ```

> Doesn't gfortran require passing the string lengths as additional parameters at the end? Don't think so

Could you compile [bugreport-arpack-dseupd_c.c](https://github.com/opencollab/arpack-ng/files/5943909/bugreport-arpack-dseupd_c-c.txt) in debug mode on mips64el and report here a backtrace? May be helpful to get some clue

In [bugreport-arpack-dseupd_c.c](https://github.com/opencollab/arpack-ng/files/5943909/bugreport-arpack-dseupd_c-c.txt), why do you set `double tol=+0x1p-53;`? This is a weird format that I never seen before (not sure to understand it), but, it sounds like `1e-53`. If so,...

1. Try to use a regular tolerance (1.e-10 or 1.e-12 - below this has no meaning and may even trigger numerical instabilities). 2. Make sure you don't miss cases where...

To me, looks more like an arpack miuse that an arpack bug. @jgmbenoit: if so, when OK at your side, don't forget to close this issue at our side. Thanks!

>Doesn't that mean that there will never be an ido == 2 value? Sounds likely, but, not sure: doc is not clear to me (when `bmat = 'I'` you have...

@szhorvat: BTW, you may want/need to make some statistics with your baseline data/use-cases. Say you ask for `tol = 1.e-12` and `maxit = 100`, but that, you observe on your...

If you don't have `ido == 2`, make sure you use a regular tolerance (1.e-12), check `maxit` and the quality of the results (are your vectors eigen up-to the tolerance...

BTW, you need to set at least `nbCV = 2 * nbEV + 1`, but, if you set `nbCV > 2 * nbEV + 1` you may get better/faster CV:...