Take advantage of variadic templates to implement interface bookkeeping and interface casting

This patch updates the implementation of the interface bookkeeping and casting from the current macros + boost mpl code to use variadic templates.

To answer my own question: No, we cannot, as we're not comparing InterfaceId<...> to InterfaceId<...> but InterfaceID to InterfaceId<...> and the conversion between those is a one way street, namely InterfaceId<...> -> InterfaceID.

What could be done (which is a variation on what is currently there) is to add some type-erasure, and require an interface to check for matches, but that will always involve runtime polymorphism. So I don't think this can ever be implemented at compile time.

Title changed from Dev variadic interface base to Use variadic templates to interface bookkeeping and interface casting

Title changed from Use variadic templates to interface bookkeeping and interface casting to Take advantage of variadic templates to implement interface bookkeeping and interface casting

Title changed from Take advantage of variadic templates to implement interface bookkeeping and interface casting to [WIP] Take advantage of variadic templates to implement interface bookkeeping and interface casting

Title changed from [WIP] Take advantage of variadic templates to implement interface bookkeeping and interface casting to Take advantage of variadic templates to implement interface bookkeeping and interface casting

LHCb validation running: https://buildlhcb.cern.ch/nightlies/index.html?slot=lhcb-gaudi-merge&build_id=4

Reassigned to @clemenci

The build fails in several places, possibly because of f6ed9be3.

Added 14 commits:

• 54783eb2...70dc41d9 - 12 commits from branch gaudi:master
• 8c91511c - Merge branch 'master' of ssh://gitlab.cern.ch:7999/gaudi/Gaudi into dev-variadic-interface-base
• 138ed5f9 - alternate fix to GAUDI-1081 was already included in this branch...

I have an alternative for f6ed9be3, namely 76739f8d which is (was) already part of this branch (but not in the modernize one -- sorry). I've merged the master branch into this branch, reverted f6ed9be3, and it compiles, and the Gaudi tests all pass

Thanks, I started a new build: https://buildlhcb.cern.ch/nightlies/index.html?slot=lhcb-gaudi-merge&build_id=6

Added 3 commits:

• e36e6a01 - 1 commit from branch gaudi:master
• 31cd7962 - Merge branch 'master', remote branch 'origin' into dev-variadic-interface-base
• 6433a2ac - restore Gaudi::Cast

With 6433a2ac included, it compiles cleanly with x86_64-slc6-gcc48-dbg as well...

Added 12 commits:

• 6433a2ac...d53e7505 - 11 commits from branch gaudi:master
• 3083c397 - Merge branch 'master' of ssh://gitlab.cern.ch:7999/gaudi/Gaudi into dev-variadic-interface-base

@clemenci could you please start a new build based on 3083c397?

You should have it soon as https://buildlhcb.cern.ch/nightlies/index.html?slot=lhcb-gaudi-merge&build_id=8

@clemenci the LHCb test fail because there is a request to add the following to a dictionary:

  <class name =        "implements1<IIncidentListener>" />
  <class name = "extend_interfaces1<IIncidentListener>" />

I'm not sure I understand why this is done: I can imagine why IIncidentLister would be added, I can imagine why some actual implementation would be added, but not why the above mixin would be included... so is there any reason to not just delete those two lines?

as an aside: it is a bit strange that cling does a SEGV when it encounters this...

Added 8 commits:

• 3083c397...850d8016 - 7 commits from branch gaudi:master
• 7c08af34 - Merge branch 'master' of ssh://gitlab.cern.ch:7999/gaudi/Gaudi into dev-variadic-interface-base

Added 24 commits:

• 7c08af34...35dd00c0 - 23 commits from branch gaudi:master
• 7c98e4d8 - Merge branch 'master' of ssh://gitlab.cern.ch:7999/gaudi/Gaudi into dev-variadic-interface-base

Added 6 commits:

• 7c98e4d8...c1d7111d - 5 commits from branch gaudi:master
• 73c1da77 - Merge branch 'master' of ssh://gitlab.cern.ch:7999/gaudi/Gaudi into dev-variadic-interface-base

Sorry for the late reply. These dictionaries were needed at the time of ROOT 5 for interactivity (see GAUDI-1098). Moreover, when I added those lines, implements1 and extend_interfaces1 were mandatory, but with some later update (if I'm not mistaken) they became unnecessary.

Just remove them.

OK, done, see https://svnweb.cern.ch/trac/lhcb/changeset/195326 -- can you please start a new build on 73c1da77?

https://buildlhcb.cern.ch/nightlies/index.html?slot=lhcb-gaudi-merge&build_id=15

lhcb-gaudi-merge#15

I've removed a few items from dictionaries in Phys, see https://svnweb.cern.ch/trac/lhcb/changeset/195356 and https://svnweb.cern.ch/trac/lhcb/changeset/195355 which should fix the Phys build.

Validation started with lhcb-gaudi-merge#16

Added 9 commits:

• 73c1da77...70dd6a76 - 8 commits from branch gaudi:master
• 8717e423 - Merge remote branch 'Gaudi/master' into dev-variadic-interface-base

Validation started with lhcb-gaudi-merge#17

Status changed to merged

mentioned in commit 328c078a

mentioned in commit 4a18175b

Mentioned in commit 775353a4