0005538: combining -i and -annot

Notes
(0007062) gasche (developer) 2012-03-14 06:27	I have a related request: I recently noticed that `ocamlc -i foo.ml` working fine does not actually provide any assurance that the code compile fines, because some static checking is performed after type-checking (in my case a "this kind of expression is not allowed in a let rec..."). So to be sure that I get all statically-detected errors, I have to use `ocamlc -c foo.ml` which is less informative. It would be nice if `ocamlc -i -c foo.ml` gave me the best of both worlds, while it currently ignores the `-c` part.

(0007064) frisch (developer) 2012-03-14 08:47	In addition to performing type-checking, "-i" also prints the inferred interface. Maybe we should introduce a new "type-check only" flag, which does the full type-checking (and produce .annot files if requested), without printing the interface.

(0007165) xleroy (administrator) 2012-03-26 15:09	Several OCaml developers are discussing this PR right now, and we're not sure what usage scenario you have in mind. Why is it that you can't use "ocamlc -c -annot" ?

(0007178) garrigue (manager) 2012-03-27 05:03	I just stumbled on it when trying to look both at the interface generated for a unit and its annotations. This was for debugging purposes, but this could happen in a developing tool too. I'm not 100% sure what is the right behavior. Note that -annot is not linked to compilation: it may output some information even if type checking fails.

(0007190) gasche (developer) 2012-03-27 10:45	I believe the expected behavior is rather clear: don't ignore any explicitly passed flag, even in presence of -i. - if -annot is passed, produce -annot files before printing the interface - if -c is passed, then compile (and report related errors) after printing the interface If there was an additional typecheck-only flag as suggested by Alain, it would be implied by -i and -annot, but would not imply interface printing -- a reasonable rationale for this is that interface printing can be slow.

(0007192) xleroy (administrator) 2012-03-27 10:52	I am not questioning the expected behavior: I'm questioning the need for yet another flag or combination of flags, given that "ocamlc -i" followed by "ocamlc -c -dannot" does the job in this case. In the past, -i was a modifier on -c and would not prevent generation of compiled files, and some users made a reasonable point that -i should print the inferred interface and do nothing else. Could we just stick with this behavior?

(0007200) doligez (manager) 2012-03-27 17:11	I have looked into the code and noticed that "ocamlopt" does allow -i and -annot together. Moreover, "ocamlc -i" also fails to check for warnings-as-errors and to remove preprocessed temporary files. I have brought ocamlc in line with ocamlopt on all these points. Until the next release, you can work around this issue by using "ocamlopt -i -annot". Fixed in 4.00 (commit 12284) and trunk (commit 12285).

Issue History
Date Modified	Username	Field	Change
2012-03-14 02:24	garrigue	New Issue
2012-03-14 06:27	gasche	Note Added: 0007062
2012-03-14 08:47	frisch	Note Added: 0007064
2012-03-26 15:09	xleroy	Note Added: 0007165
2012-03-26 15:09	xleroy	Status	new => feedback
2012-03-27 05:03	garrigue	Note Added: 0007178
2012-03-27 05:03	garrigue	Status	feedback => new
2012-03-27 10:45	gasche	Note Added: 0007190
2012-03-27 10:52	xleroy	Note Added: 0007192
2012-03-27 10:52	xleroy	Status	new => resolved
2012-03-27 10:52	xleroy	Resolution	open => suspended
2012-03-27 17:11	doligez	Note Added: 0007200
2012-03-27 17:11	doligez	Status	resolved => closed
2012-03-27 17:11	doligez	Resolution	suspended => fixed
2012-03-27 17:11	doligez	Fixed in Version	=> 4.00.0+dev

Relationships