some issues with hasCustomPragma/recordFields

[jangko]

1. getType(T)[1].getImpl will not work on ref type including deref version like type(default(someref)[]).

possible work around. after getType, use new owner API then symKind and typeKind to check if it has parent type or something else.

nim-confutils, nim-serialization, and nim-json-serialization suffer from this problem too.

2. nnkPragma and nnkPosfix in recordFields

          if field.name.kind == nnkPragmaExpr:
            field.pragmas = field.name[1]
            field.name = field.name[0]

          if field.name.kind == nnkPostfix:
            field.isPublic = true
            field.name = field.name[1]

this is clearly wrong, nnkPostFix will appear inside nnkPragmaExpr in the AST. need fix

3. using nnkRecWhen in recordFields can leads to unexpected result. should use semchecked AST of getType instead of raw AST from getImpl.

nnkRecWhen will not appear at final type, only branch evaluated to true will appear at final type. if we use nnkRecWhen, both false branch and true branch will have conflict if they contains fields with same name. false branch existence also posses it's own problem if we count the field number or enumerate the fields.

work around: not known, getType and friends return an AST without pragmas. need to patch the getType family.

related: nim-lang/RFCs#176

[zah]

this is clearly wrong, nnkPostfix will appear inside nnkPragmaExpr in the AST. need fix

But that's what the code is doing now. It first strips the "outer" layer, which is the nnkPragmaExpr, then moves on to strip the inner one. I believe this is code path is already tested in confutils.

3. The nnkRecWhen problem is a tough one. What I really need is the Nim compiler to offer an option to handle this for me and to return the fields in the record after when was applied.

[jangko]

But that's what the code is doing now. It first strips the "outer" layer, which is the nnkPragmaExpr, then moves on to strip the inner one. I believe this is code path is already tested in confutils.

sorry, you're right.I must be confused with something else

[jlokier]

@zah:

The nnkRecWhen problem is a tough one. What I really need is the Nim compiler to offer an option to handle this for me and to return the fields in the record after when was applied.

I have just encountered this when issue after adding a confutils option to nimbus-eth1 that should only be present in some build configurations. We didn't have this problem in the past, but now we've switched to confutils we do.

Is there a way to make when work nowadays? The conditional is a very simple constant, when defined(evmc_enabled), and the options list it fits into is long (and there will be more conditionally-compiled options I expect).

[jangko]

@jlokier: now nim-confutils already support ignore pragma in status-im/nim-confutils#37, the closest workaround to achieve conditional fields is to use pragma-pragma