[SR-6736] Enforce 16-bit limit for # of function parameters, # of tuple type element

[DougGregor]

Previous ID	SR-6736
Radar	None
Original Reporter	@DougGregor
Type	Bug

Additional Detail from JIRA

Votes	0
Component/s	Compiler
Labels	Bug, StarterBug
Assignee	None
Priority	Medium

md5: ea27fac0ee252473c9618a9257498092

Issue Description:

Swift's ABI uses 16 bits to store the number of parameter types in a function type and the number of element types in a tuple type. We should enforce these limits in the parser, so they don't slip through and become crashes at later phases.

[swift-ci]

Comment by Yasser Deceukelier (JIRA)

I might be interested in taking a look at this one.

But before I do: how are the number of arguments represented, unsigned binary (i.e. max 2^16 arguments), or is it signed? Are all the bits used to represent the number of arguments, or are some reserved?

[DougGregor]

We have the full 16 bits stored as an unsigned value in the metadata.

[davezarzycki]

Personally speaking, 4 bits should be enough here. If you need more than 16 parameters or tuple elements, then you really should be using a `struct`. That being said, some people enjoy complexity and therefore 8 bits should conservatively split the difference between "reasonable" code and "unreasonable" code.

[swift-ci]

Comment by Yasser Deceukelier (JIRA)

Ok, got it.

I'm new to the swift compiler, so I might need to ask some additional questions.

Enforcing this limit, should it be done in the parser or can it happen as a check on the AST? (The latter sounds better in my opinion.) And if it could be done in the AST, which source file could I use as a reference/starting point to search where to add this feature?

[davezarzycki]

At the very least, start by auditing the various AST verifiers in lib/AST/ASTVerifier.cpp. From there, you should gain a good view on what data structures are involved. Next, you can work towards having Sema's type checking diagnose when too many parameters / tuple elements are seen.

Given various other goals, I doubt the AST will ever be optimized to the point where it cannot represent invalid code, therefore I don't think parsing will ever need to diagnose when limits (like the number of parameters) are exceeded.

[belkadan]

Resetting assignee for all Starter Bugs last modified in 2018.