[BUG FIX - architecture] fix arpackdef.h: must not be included by F77/F90.

[fghoussen]

This should fail.

[coveralls]

Pull Request Test Coverage Report for Build 1145

• 0 of 0 changed or added relevant lines in 0 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage remained the same at 73.187%

---

Totals
Change from base Build 1125:	0.0%
Covered Lines:	16462
Relevant Lines:	22493

---

💛 - Coveralls

[fghoussen]

So this is a noisy bug which fails with :

../../arpackdef.h:8:18: error: variable or field 'cnaupd_c' declared void
 #define a_int    int64_t
                  ^~~~~~~

But, this can be fixed... And likely end-up into a later (runtime) failure that should look suspiciously like #230...

[fghoussen]

So here as "expected", tests fail on :

FAIL: icb_arpack_cpp
====================
ERROR: iparam[4] 0, nev 9, info -1
terminate called after throwing an instance of 'std::domain_error'
  what():  Error inside ARPACK routines
FAIL icb_arpack_cpp (exit status: 134)

But you can keep this fix if you make "disappear" the problem reverting 489eb89.

[fghoussen]

Ah, OK... Bad news: this is a casper bug on ILP64... OK on master with --enable-icb here

arpack-ng/.travis.yml

Line 268 in cfdb5f6

./configure --with-blas=mkl_gf_ilp64 --with-lapack=mkl_gf_ilp64 --enable-icb --enable-mpi \

, KO after af30be6 with --enable-icb-exmm which implies --enable-icb , and, finally reverting 0e43bce does not get back to master behavior as it should. So it seems the problem may or not occur on ILP64 depending on runs. The only way out for now is to mute the bug on ILP64.

[fghoussen]

This one is a real bug nest... So I'll restart all this from scratch

[fghoussen]

Let's restart this crazy casper-bug story from the start...

[fghoussen]

Step 1 : on master, currently on ILP64 CI build you have --enable-icb here

arpack-ng/.travis.yml

Line 268 in cfdb5f6

./configure --with-blas=mkl_gf_ilp64 --with-lapack=mkl_gf_ilp64 --enable-icb --enable-mpi \

and this enable switch triggers icb_* tests : CI is OK, which neams "icb" tests passed (https://travis-ci.org/opencollab/arpack-ng/jobs/655313079?utm_medium=notification&utm_source=github_status)

[fghoussen]

Step 2 : on this PR, on ILP64 CI build, my first commit is meant to replace --enable-icb by --eanble-icb-exmm which will do more things, that is, running "icb" tests and also "arpackmm" tests (which need ICB) which are not yet run on master.

This (5fbcb67) will fail this way :

../../arpackdef.h:8:18: error: variable or field 'cnaupd_c' declared void
 #define a_int    int64_t
                  ^~~~~~~

[fghoussen]

OK, we get the ILP64 compilation problem we expected.

Step 3 : fix only compilation. Tests should run exactly like before (pass or fail). Tests passed before... But casper may make them fail or pass now... Let's see.

[fghoussen]

Step 4 : say hello to casper who has make icb tests fail !?...

FAIL: icb_arpack_c
==================
FAIL icb_arpack_c (exit status: 139)

FAIL: icb_arpack_cpp
====================
ERROR: iparam[4] 0, nev 9, info -1
terminate called after throwing an instance of 'std::domain_error'
  what():  Error inside ARPACK routines
FAIL icb_arpack_cpp (exit status: 134)

[fghoussen]

Step 5 : (try to) say goodbye to casper who may not want to go...
Reverting with 6768195 must get us back to master : I seem this may not happen... Let's see

[fghoussen]

Casper'ed again: we still have a problem while we should not have it (CI is back as it is on master)

FAIL: icb_arpack_c
==================
FAIL icb_arpack_c (exit status: 139)

FAIL: icb_arpack_cpp
====================
ERROR: iparam[4] 0, nev 9, info -3
terminate called after throwing an instance of 'std::domain_error'
  what():  Error inside ARPACK routines
FAIL icb_arpack_cpp (exit status: 134)

[fghoussen]

Step 6 : the only way out is to call ghost-bust-bored (0b4ab15).... Until the problem is understood.

[fghoussen]

This is a real problem. Must be merged.

[fghoussen]

Must be merged first !

[fghoussen]

Now included in #249 (first step of it).