Make sure that recursive dependencies are not reflexive

[LasseBlaauwbroek]

opam list --recursive --recursive --required-by=package is reflexive (it shows the package itself). This does not seem correct, since a package does not require itself. Note that without --recursive, the command is not reflexive.

[kit-ty-kate]

Thanks! fixes #4446

[kit-ty-kate]

The tests fail and it might be related. Could you have a look at them? e.g.

diff --git a/tests/reftests/legacy-git.test b/tests/reftests/legacy-git.out
index a629f71..63d38d2 100644
--- a/tests/reftests/legacy-git.test
+++ b/tests/reftests/legacy-git.out
@@ -747,20 +747,16 @@ Done.
 ocaml.system
 P1.1
 ### opam remove P1
-The following actions will be performed:
-  - remove P1 1
-
-<><> Processing actions <><><><><><><><><><><><><><><><><><><><><><><><><><><><>
--> removed   P1.1
-Done.
+Nothing to do.
 ### opam list -is --columns=package
 ocaml.system
+P1.1
 ### : LIST :
 ### opam list -A
 # Packages matching: any
 # Name # Installed # Synopsis
 ocaml  system
-P1     --          A very useful package
+P1     1           A very useful package
 P2     --          An other very useful package
 P3     --          Testing version names
 P4     --          Testing transitive closure

[LasseBlaauwbroek]

My guess is that some functions rely on the reflexive behavior (for example, in order to calculate package removal). I changed it so that the non-reflexive behavior only occurs in the list command.

[AltGr]

Thanks! Indeed this was inconsistent.

A note of caution however: this is incorrect with multiple packages, e.g. --required-by=a,b should still return a if b depends on a.

[LasseBlaauwbroek]

I guess this is true, although it depends a bit on how one reads the documentation of --required-by and --depends-on.

The only way I can think of fixing this is to run a separate query for each of the requested atoms and then take the union of the results. Would that be acceptable, or is that too slow?

[AltGr]

I guess this is true, although it depends a bit on how one reads the documentation of --required-by and --depends-on.

Hm, that doc probably needs some effort then. The idea is that --depends foo,bar is not the same as --depends foo --depends bar, because the latter wouldn't need an extra syntax.

It's not a very common pattern, so I guess the solution you propose would be fine, dependency queries are not that expensive; and I can't think of an other one :)

[kit-ty-kate]

Superseded by #5075 which supports the --required-by=a,b case.
Thanks a lot for getting the ball rolling!