how to crash REPLICATE ?

[alaingdl]

z=replicate(3,!values.f_nan)

z=replicate(3,!values.f_infinity) is OK

OK, I made a typo 🥲 (interverting first and second arg.)

(I am not sure to be able to look at that today)

[alaingdl]

What I was afraid of is confirmed 😢

IDL> B = BYTARR(2, !values.f_nan)
% Array dimensions must be greater than 0.

versus

GDL> B = BYTARR(2, !values.f_nan)
GDL> help, b
B               BYTE      = Array[2, 9223372036854775808]

We do face a serious problem, protecting all the inputs

[alaingdl]

identity( !values.f_nan) and z=findgen(3,!values.f_nan) crashed too ...

[jtappin]

IDL> B = BYTARR(2, !values.f_nan)
% Array dimensions must be greater than 0.

Isn't exactly suggestive of elegant handling of the case in IDL either.
A quick investigation suggests that when encountering a floating point dimension argument IDL computes LONG64, while GDL computes ULONG64. And long64(!values.f_nan) = -9223372036854775808 while
ulong64(!values.f_nan) = 9223372036854775808.

I don't think that's really the proper way to handle the situation though.

[GillesDuvert]

I believe we should use LONG64 .

[alaingdl]

I think we have an opportunity to deliver a clever message than IDL

It is not urgent, less urgent that "operational" bugs or regressions. Then we can think on solutions.
(if e->ParamIsNotFinite(2) then ...)

[GillesDuvert]

the patch will be sufficient for the time being I believe.