0005411: New directive for the toplevel: #load_rec

(0006270)
frisch (developer)
2011-12-13 17:43

I believe the proposed patch does not work when the name of the module contains uppercase characters (because of the call to String.lowercase).

From an implementation point of view, it would be cleaner to iterate through compunit.cu_reloc in load_file (in 'rec' mode) in order to load missing modules (and not waiting for an exception to pop up).

That said, I'm not 100% convinced by the proposal: a module can be provided by a .cma file, and there is no easy way to know which one. Do you have a convincing argument in favor of your suggestion? What would be a typical use case?

(0006273)
dim (developer)
2011-12-13 18:08

> I believe the proposed patch does not work when the name of the module contains uppercase characters (because of the call to String.lowercase).

Indeed. It is a typo, it should be String.uncapitalize.

> That said, I'm not 100% convinced by the proposal: a module can be provided by a .cma file, and there is no easy way to know which one. Do you have a convincing argument in favor of your suggestion? What would be a typical use case?

The typical use is the following: you have a project with a lot of modules and you want to test in the toplevel an internal function of one of them. First you load dependencies with #require, then you have to load all the cmos of you project needed by the function you want to test, and you have to load them in the right order.

In general you do that:

# load "a.cmo";;
Error: Reference to undefined global `B'
# load "b.cmo";;
Error: Reference to undefined global `C'
# load "c.cmo";;
# load "b.cmo";;
# load "a.cmo";;

The goal of this directive is just to do this mechanical stuff automatically.

(0006275)
mehdi (reporter)
2011-12-13 18:47

You can also put these directives in a file (say "foo.ml") and then, in a toplevel, do “#use "foo.ml";;”. You'll need to resolve the dependencies once to start, and update the file when needed.

You can also implement your proposal in a "foo.ml" and #use it.

(0006276)
frisch (developer)
2011-12-13 18:51

Ok, wish granted. I'm a little bit uneasy with the fact that the loading order of dependent .cmo files is not specified (and this can change the semantics), but I can see there are cases where this can be useful.

Committed to the trunk (rev 11306). Can you try it?

Note that the recursive behavior is only applied when .cmo files (not .cma) are loaded.

(0006277)
dim (developer)
2011-12-13 19:11

> Committed to the trunk (rev 11306). Can you try it?

Yes, it works. Thanks.

> Note that the recursive behavior is only applied when .cmo files (not .cma) are loaded.

OK.

By the way, i just noticed that load_file does not close the input channel in case an exception (other than Load_failed) is raised. So if for example "#load "a.cmo";;" fails because of a reference to an undefined global the file will be kept open.

(0006288)
frisch (developer)
2011-12-14 12:04

Thanks, I've fixed load_file to close the input channel in case of error (commit 11315).

Issue History
Date Modified	Username	Field	Change
2011-11-30 10:47	dim	New Issue
2011-11-30 10:47	dim	File Added: dir-load_rec.diff
2011-12-13 17:43	frisch	Note Added: 0006270
2011-12-13 18:08	dim	Note Added: 0006273
2011-12-13 18:47	mehdi	Note Added: 0006275
2011-12-13 18:51	frisch	Note Added: 0006276
2011-12-13 18:51	frisch	Assigned To	=> frisch
2011-12-13 18:51	frisch	Status	new => assigned
2011-12-13 18:52	frisch	Status	assigned => feedback
2011-12-13 19:11	dim	Note Added: 0006277
2011-12-13 19:11	dim	Status	feedback => assigned
2011-12-14 12:04	frisch	Note Added: 0006288
2011-12-14 12:04	frisch	Status	assigned => resolved
2011-12-14 12:04	frisch	Resolution	open => fixed
2012-09-25 20:07	xleroy	Status	resolved => closed

Notes
(0006270) frisch (developer) 2011-12-13 17:43	I believe the proposed patch does not work when the name of the module contains uppercase characters (because of the call to String.lowercase). From an implementation point of view, it would be cleaner to iterate through compunit.cu_reloc in load_file (in 'rec' mode) in order to load missing modules (and not waiting for an exception to pop up). That said, I'm not 100% convinced by the proposal: a module can be provided by a .cma file, and there is no easy way to know which one. Do you have a convincing argument in favor of your suggestion? What would be a typical use case?

(0006273) dim (developer) 2011-12-13 18:08	> I believe the proposed patch does not work when the name of the module contains uppercase characters (because of the call to String.lowercase). Indeed. It is a typo, it should be String.uncapitalize. > That said, I'm not 100% convinced by the proposal: a module can be provided by a .cma file, and there is no easy way to know which one. Do you have a convincing argument in favor of your suggestion? What would be a typical use case? The typical use is the following: you have a project with a lot of modules and you want to test in the toplevel an internal function of one of them. First you load dependencies with #require, then you have to load all the cmos of you project needed by the function you want to test, and you have to load them in the right order. In general you do that: # load "a.cmo";; Error: Reference to undefined global `B' # load "b.cmo";; Error: Reference to undefined global `C' # load "c.cmo";; # load "b.cmo";; # load "a.cmo";; The goal of this directive is just to do this mechanical stuff automatically.

(0006275) mehdi (reporter) 2011-12-13 18:47	You can also put these directives in a file (say "foo.ml") and then, in a toplevel, do “#use "foo.ml";;”. You'll need to resolve the dependencies once to start, and update the file when needed. You can also implement your proposal in a "foo.ml" and #use it.

(0006276) frisch (developer) 2011-12-13 18:51	Ok, wish granted. I'm a little bit uneasy with the fact that the loading order of dependent .cmo files is not specified (and this can change the semantics), but I can see there are cases where this can be useful. Committed to the trunk (rev 11306). Can you try it? Note that the recursive behavior is only applied when .cmo files (not .cma) are loaded.

(0006277) dim (developer) 2011-12-13 19:11	> Committed to the trunk (rev 11306). Can you try it? Yes, it works. Thanks. > Note that the recursive behavior is only applied when .cmo files (not .cma) are loaded. OK. By the way, i just noticed that load_file does not close the input channel in case an exception (other than Load_failed) is raised. So if for example "#load "a.cmo";;" fails because of a reference to an undefined global the file will be kept open.

(0006288) frisch (developer) 2011-12-14 12:04	Thanks, I've fixed load_file to close the input channel in case of error (commit 11315).

Relationships