fixed tests (failures are not related to this pull request)

I've been hit by the bug, patch fixes the issue 👍

I would have preferred to actually inline the service instead of keeping it as a separate method.

hmmm, as this breaks the XML dump, we need a test for the XML dumper, to ensure we don't reintroduce the bug later on.

@fabpot What kind of behaviour would you expect in theXmlDumper when it gets passed an "inlined" service factory? Should it throw an exception?

XML-dumping a definition using a private factory fails without your patch, and works with. This could be tested.

I suggest adding a test taking each of our container fixture in the testsuite, compiling the container, and then dumping it as XML. this way, we would detect cases whichcannot be dumped

@nicolas-grekas The issue with your suggestion is that this is no related to theXmlDumper itself, but is related to how the instance looks like that is passed to it (i.e. the container as it has been modified by all compiler passes). So I added a test for theInlineServiceDefinitionPass which ensures that factories are not inlined.

@stof Doing so reveals that theXmlDumper also isn't capable of dumping inlined configurators (I suggested a way to dump them in#13557). Should I remove this possibility here too (given that adding such an option might be considered a new feature)?

By the way, most of the container fixtures were not usable because of undefined parameters. I added tests for the others though as you suggested.

Thank you@xabbuh.

@fabpot Any opinion on the dumper (see the PR description)? I tend to adapt it as well in case someone uses the dumper without the parameter resolver pass (parameters in all other features seem to be resolved there too). What do you think?