[MLton] bug report, vector of char problem

Daniel C. Wang danwang@CS.Princeton.EDU
Wed, 06 Apr 2005 15:12:03 -0400

In anycase, I don't disagree with the premises that typed-ILs are a good 
thing. Just wondering if there is some "easy" to dig up evidence that might 
make a skeptical person less so... something beyond anecdotal experience.

Here's a simple number to generate. Every time the internal type-checker 
fails it emails or logs the failure some place. One would hope that when 
developing a new optimization the type-checker fails a few times. If the 
type-checker never fails during normal development of the compiler then we

Every time it does fail to type-check it is identifying a real bug. So we 
count that as a success for the typed-IL. Anyway, just some food for thought.

Stephen Weeks wrote:
>>>However, the changelog is not a good indicator of how helpful typed
>>>ILs are, as most of their use comes when developing a new pass, while
>>>changes and debugging is fast and furious, before anything makes it
>>>into the changelog.
>>I agree with this completely.  That's why I was stating that the changelog 
>>would be a source of the examples Dan was looking for: bugs that didn't 
>>trickle a bug in the typed-IL.
> Agreed.  I was confused.
> Although, just because a bug makes it out the door into a release
> doesn't mean that the type checker wouldn't catch it.  The first thing
> I always do when we get a bug report is compile with -type-check
> true.