Linux wheels builds failing with: error: lvalue required as unary ‘&’ operand

All the Linux wheels builds have been failing for some time, looks like since !1608 (merged) was merged.

There are many errors of the form:

LALSimIMRPhenomTHM_internals.c: In function ‘IMRPhenomTSetPhase22Coefficients’:
LALSimIMRPhenomTHM_internals.c:408:16: error: lvalue required as unary ‘&’ operand
  408 |  REAL8 ascut = GSL_REAL(gsl_complex_arcsinh(phi));
      |                ^~~~~~~~

Full build log is here.

/cc @riccardo-sturani, @@leo-singer

cc @hector.estelles as PhenomTHM developer

added lalsimulation prioritymustfix labels

The wheels are currently built with GSL 2.6. I noticed the following in the GSL 2.7 change log:

What is new in gsl-2.7:

...

• added support for native C complex number types in gsl_complex when using a C11 compiler

I'll try updating GSL. Does LALSuite now require GSL 2.7? If so, then the check in configure.ac should be updated to reflect that.

Wait, no, I already updated to GSL 2.7 in lalsuite-manylinux!20 (merged)...

The date that the wheels started failing is also consistent with the date that I merged lalsuite-manylinux!20 (merged). So does LALSuite require GSL <= 2.6?

The macOS machines have GSL-2.7 and don't exhibit the same error.

Can you please look at this ASAP as I'm trying to get a release out. If it's not addressed by the end of the week I will revert the change that introduced the problem.

@leo-singer, @hector.estelles, I think I have seen that make changes of this sort resolves the errors:

@@ -1753,7 +1755,8 @@ double IMRPhenomTRDAmpAnsatzHM(REAL8 t, IMRPhenomTHMAmpStruct *pAmp)
   REAL8 c1 = pAmp->c1_prec;

   gsl_complex phi = gsl_complex_rect(c2*(t-tpeak) + c3,0);
-  REAL8 tanh = GSL_REAL(gsl_complex_tanh(phi));
+  gsl_complex tanhc = gsl_complex_tanh(phi);
+  REAL8 tanh = GSL_REAL(tanhc);

   REAL8 expAlpha = exp(-pAmp->alpha1RD_prec*(t-tpeak));

but I have no real understand of why this works.

For (fairly) obvious reasons, the following works, and may be cleaner:

@@ -1752,8 +1759,8 @@ double IMRPhenomTRDAmpAnsatzHM(REAL8 t, IMRPhenomTHMAmpStruct *pAmp)
   REAL8 c2 = pAmp->c2_prec;
   REAL8 c1 = pAmp->c1_prec;

-  gsl_complex phi = gsl_complex_rect(c2*(t-tpeak) + c3,0);
-  REAL8 tanh = GSL_REAL(gsl_complex_tanh(phi));
+  gsl_complex phi = gsl_complex_tanh(gsl_complex_rect(c2*(t-tpeak) + c3,0));
+  REAL8 tanh = GSL_REAL(phi);

   REAL8 expAlpha = exp(-pAmp->alpha1RD_prec*(t-tpeak));

Does GSL_REAL((gsl_complex) gsl_complex_arcsinh(phi)) work?

@leo-singer, no:

LALSimIMRPhenomTHM_internals.c: In function 'IMRPhenomTRDAmpAnsatzHM':
LALSimIMRPhenomTHM_internals.c:1756:16: error: lvalue required as unary '&' operand
 1756 |   REAL8 tanh = GSL_REAL((gsl_complex) gsl_complex_tanh(phi));
      |                ^~~~~~~~

mentioned in merge request !1632 (merged)

mentioned in merge request !1631 (merged)

mentioned in merge request !1634 (merged)

assigned to @duncanmmacleod

mentioned in commit 103b3bf2

closed with merge request !1634 (merged)

mentioned in merge request gstlal!104 (merged)