fingolfin
commented
3 days ago Thank you!
This behaviour of recog is not nice, the people who invented it did probably not think of the situation that the package gets installed automatically by PackageManager
They absolutely did not, because when I added regen_doc.g in 2017, PackageManager did not even exist :-). And to this day nobody apparently tried to install recog via PackageManager either (or at least they did not bother to write an issue for recog), otherwise "we" would have looked into fixing it long ago. (Though looking at the changes you made, they seem to me more like a workaround for a bug in PackageManager than a fix? Note that makedoc.g in virtually all GAP packages really expects to be run from the directory it lives in; that it apparently often works otherwise as well is just luck (whether good or bad luck is a matter of opinion, after all it hid the bug in PackageManager for a long time)
codecov[bot]
Return to the situation where the released version of GAP's
recogpackage gets installed if necessary, not a current version from github.We keep the change that the
OscarInterfacepackage gets loaded in the end, and that it requires thePolycyclicpackage.I think that once there will be a new released
recogversion, theinstallcall forrecogin Oscar will run into an error unless we are then able to start GAP without loaded packages. The point is that if a version is already loaded that lies in a path different from the path of the newly installed version, the GAP code inrecogthat gets called for processing the documentation will throw an error. This behaviour ofrecogis not nice, the people who invented it did probably not think of the situation that the package gets installed automatically byPackageManager.