* 

* Add a README in experiments/CCompiler, including where to find
  licence information, software requirements: menhir, ocaml, CIL
  (included in distribution)...  how to compile how to run and test

* Put trace of development in a separate directory outside the
  distribution.

* In the sources in experiments/backend/src: put Annot_all (direct
  approach) outside the distribution trace origin/licence of each file


* It would be good to separate as much as possible the basic compiler
  from the part which has to do with the annotations (labelling,
  instrumentation, annotation, computation cost on the control flow
  graph).  Parts that need to be kept in the same file should be
  clearly marked.  Ideally, it should be possible to extract
  quasi-automatically the distribution for the basic compiler.

* Increase readability of source code. Currently there are too many
  files.  For instance, create a directory for each intermediate
  language and a separate directory for shared algorithms, memory
  model.

* Clarify what part of C is covered. Currently, wrt to CompCert we
  have no floating point and no dynamic allocation of arrays. Moreover
  starting from RTL_MIPS we do not cover all C data formats. Anything
  else ?

  => Currently, local arrays can be handled through allocation in the
  heap.

* Prepare battery of tests. Simple tests, CompCert tests, programs
  generated by Lustre compiler. When looking at this tests, we should
  identify the part of C that we definitely want to be able to compile.

* In the commands, we could add the possibility to select the target
  assembly language (currently only MIPS, but ARM7, 8051 in the
  future).

* Clarify parts of the compiler that are sensitive to retargetting.
  For instance, what if we move to 8 bits integers (8051) ?
  => Look into SDCC manual.

* Possibility to define an interpreter for subset of ARM7, 8051 used
  in the compilation. Necessity to fix simulators for ARM7, 8051.