On Fri, 2006-01-06 at 10:10 -0500, Eric Cooper wrote: > This isn't a substitute for a real test suite, but what if the Debian > OCaml policy were to say that all build rules should use "-warn-error A" > (or "-warn-error As" for lablgtk apps). That way, at least new > warnings from new compiler releases would force a FTBFS.
That is too strict and probably impossible to enforce. First: My compiles use -w yzex -warn-error FDPSU at the moment. I cannot and do not want to error out code that choses to name unused function parameters, for example: one could even say that in the presence of type inference this is a good thing. In other cases a named argument is used or not used depending on debugging statements which may or may not be commented out at any one time. I need to suppress even the warnings, since they clutter up the output and prevent seeing what I consider more serious. Secondly, I have third party code in my package, including FrontC/CIL and there are obvious reasons why the modifications I make to such code should be minimised. In fact I have tried to fix it so statements not returning unit are properly ignore()-ed because I consider it a serious error not to: I consider it a type error. But unused arguments are not this kind of error -- in my opinion. And I can't easily change FrontC/Cil to remove all those warnings (each one makes a diff against the latest version harder to read). And finally .. there is no way the maintainer can fix my compile commands. They're not in a makefile, they're buried deep inside a Python script, and they're constructed by consulting configuration options. Having said all that .. I do basically agree with Eric's idea: it's a proposal which attempts to upgrade quality control and get stricter about what's acceptable and what isn't. -- John Skaller <skaller at users dot sf dot net> Felix, successor to C++: http://felix.sf.net -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
