fix #6472 Typed template arguments are substituted directly if they a…

[bung87]

…re compatible with parameter type

fix #6472

after CI fails it seems need fix exists code for this.

[Araq]

I don't understand this fix.

[bung87]

have you check the issue? before this arguments will pass template's arguments type check, however when construct array will throw type mismatch. which means it passed as int.

[bung87]

also mentioned here #18667 (comment)

[metagn]

While this looks like a hack at first glance, it's a special cased version of doing argNode = semExpr(argNode, expectedType = paramType) for integers. The issue is a consequence of fitNode not being called on template parameters.

Maybe this is too special cased, and I'm not sure if it works with varargs. Since this breaks code, it could be made an experimental feature, but I'm not sure who would opt in to it.

[bung87]

While this looks like a hack at first glance, it's a special cased version of doing argNode = semExpr(argNode, expectedType = paramType) for integers. The issue is a consequence of fitNode not being called on template parameters.

Maybe this is too special cased, and I'm not sure if it works with varargs. Since this breaks code, it could be made an experimental feature, but I'm not sure who would opt in to it.

I dont think it's "special cased" or "made an experimental feature", it break packages code that depends on odd behavior, that can be fixed in user code, the problem is obvious to me, it doesn't respect concret param type user specified. I think whether it fixed like this or in other way.

[bung87]

template arguments not check right now, it may do check someday, at that time no need handle this case. so I'll close now.