[BUG FIX - ICB] iparam/ipntr sizes may change depending on cases.

[fghoussen]

And this is a silent bug...

[sylvestre]

Could you please add a test which reproduce the bug and make sure it doesn't happen anymore?
Also please update the changelog
thanks

[coveralls]

Pull Request Test Coverage Report for Build 1165

• 0 of 0 changed or added relevant lines in 0 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.8%) to 74.08%

---

Totals
Change from base Build 1158:	0.8%
Covered Lines:	16462
Relevant Lines:	22222

---

💛 - Coveralls

[fghoussen]

Could you please add a test which reproduce the bug and make sure it doesn't happen anymore?

Not really. Always / often happening : allocating more memory than needed (silent-not-hurting bug). The real question is : if 7d341a9 is OK (on LP64), then, could this fix #230 which seems to show up only on ILP64 ? (fix is related to integer )

[fghoussen]

All OK... So one can try to make the problem "reappear" here after rebase on #246 to benefit from c170c16 and check out if this could fix what looks like #230 ?

You may not want/need to merge this "bug" (allocating more memory than needed) unless it fixes ILP64 related problems.

[fghoussen]

OK, here : this is a silent-no-hurting bug. This should not create any problem : it's just that you may sometimes allocate a bit more memory than you need (14/11 integers instead of 11/7). You always allocate more : so there should be no worry here.

At first, I suspected this may be the cause of #230 (seems to be ILP64 specific) as it's related to iparam / ipntr which are arrays of integer. Seems my suspicion was wrong...

I let this PR aside for now. No problem if not merged.

[fghoussen]

After looking back at this, this should preferably be merged :

• I realized that in many cases ipntr size is actually increased in icb*.F90 ! (need more memory than expected)

• But in icb*.F90 , the sizes are likely muted variables / not-used information (not sure) and are probably "just not checked anyway" (what is checked by compiler is the type, not sure if the size is checked too)

• In pratical cases, using the same size for iparam / ipntr is more convenient (even when you allocate more than needed) : hunting all cases in the repo to "fix" this would be endless and error-prone... So I don't want to go into this. But maybe you could keep the "fix" in arpackSolver.hpp (as this one is easy : this is the one place where you always pass by)

[fghoussen]

Rebasing on master as we know now this is no big bug and can't be related to ILP64 problems.

[sylvestre]

Suggested change

	return 0;
	}
	else if (aeupd == "eupd") {
	return 0;
	}
	if (aeupd == "eupd") {

The else is useless as we have a return just above

[fghoussen]

Sure

[sylvestre]

fyi, you could just have press the "commit suggestion" button ;)

[fghoussen]

OK, now I know!