(cherry picked from commit c7caadda0902a0b5cf7194c584ffd3cc328827a1)

(cherry picked from commit 9a9e0a04822c237b1da303688293c2600bd53161)

(cherry picked from commit 04de3a90cddb2718dc263ee104854b5395bfd869)

(cherry picked from commit 0b7973cecb915abd7e0add8b907cec6784a30313)

This is checked explicitly by store/compare state.

Shouldn't matter for codegen.

*Merge PreVisitDecl and PreVisitMacro and rename to CollectFilesToUncache.
*Add FID to the transaction storing the memory buffer file ID caused the
transaction.

cleanup of offset calculations by always using (derived, base) in the interfaces and by explicit naming; also removed a workaround, and this closes ROOT-5781

the rootmapfile was created within the TSystem::CompileMacro function
assuming the "old" rootmapfile format. This upgrade allows to have
a single producer of rootmapfiles: rootcling.

This avoid picking up by mistake the content of sub-directory that
are, for example, separate out-of-source builds (e.g. with different
compiler version or optimization level)

given that dictionary headers are not created anymore since ~1month

To achieve this goal the following changes were applied:
1) rlibmap is reduced to a no-op with a nice deprecation message.
The return value is zero not to break build systems which may
rely on it. This will be transformed to an error in beta 2.
rlibmap will be removed in beta 3.
2) The build systems were updated in order to leverage rootcling
for the rootmap generation, basically adding 2 flags, -rmf and -rml.
3) Rootcling has a new option, -r which is on the same footing of
-f but it tells the system to ignore the dictionary generation.
When -r is present rootcling becomes a tool to generate rootmaps
(and capability files).

At the moment the makefile build does not properly account for
rootmaps as targets, i.e. a deleted/modified rootmap will trigger
a no-op rlibmap. A forthcoming commit will fix this behaviour.