(legacy Individual version)

ROUZAUD-CORNABAS Jonathan authored 4 years ago and TURPIN Laurent committed 4 years ago

Naming has changed, so merge was not straighforward, could contain errors.

This commit re-adds some modifications from the aevol_6 branch that
had not been kept by the large merge between the aevol_6 and aevol_7
branches.

The name SIMD was meant to specify the Single Instruction Multiple Data implementation i.e. vectorization but it is now the aevol_7 (previously known as parallel_6) version.

Tree/lineage/ancestor stats are working for both vanilla and SIMD (but reproduceability issues between both)

Magic command for reference:
find ! -path '*test*' ! -path './libaevol/SFMT-src-1.4/*' \( -name '*.cpp' -o -name '*.h' \) -print0 | xargs -0 sed -i -r 's/\bget_(\w+)\b/\1/g'

Looking forward to honour Google Style guide, the purpose is to rename getters
from get_var() to var(). Many of such renames cause conflicts, though. In this
commit, only the safe names were changed.

grep -rEo '\bget_\w+\(' --include '*.h' --no-filename --exclude-dir tests \
 | sed -e 's/get_//' -e 's/($//' \
 | sort -u \
 | xargs -n 1 -I% sh -c "grep -roq '\b%\b' --include '*.cpp' --include '*.h' --exclude-dir tests || echo %" \
 | xargs -n1 -I% sh -c "find ! -path '*test*' ! -path './libaevol/SFMT-src-1.4/*' \( -name '*.cpp' -o -name '*.h' \) -print0 \
 | xargs -0 sed -i -r 's/\bget_%\(/%\(/g'"

Yes. I know...

http://google-styleguide.googlecode.com/svn/trunk/cppguide.html#Function_Names

Functions with empty argument list should be declared f() rather than f(void).

Took aevol_run.cpp's header as reference header and set it for all Aevol source files.

* Find member variables that could clash

: grep -rEo --no-filename '\w+' --include '*.cpp' --include '*.h' | sort -u | grep '^_' > _litigious_tokens.txt
: grep -rEo --no-filename '\w+' --include '*.cpp' --include '*.h' | sort -u | grep '_$' > litigious_tokens_.txt
: grep -Fx -f <(sed -e 's/_$//' litigious_tokens_.txt) <(sed -e 's/^_//' _litigious_tokens.txt)
: cat clashes.txt
: | xargs -n 1 -I% sh -c "echo --- % --- ; grep -rlE --include '*.cpp' --include '*.h' '\W_%'
: | xargs grep -l '%_\W'"

From the resulting list,
files of interest, trimmed:
ExpManager.h | t_end
GeneticUnit.cpp | dna rna_list

False positives removed (needed no action)
Dna.cpp | indiv
ExpManager_X11.cpp | win
GeneticUnit.cpp | non_coding_computed
GeneticUnit.cpp | transcribed
Individual.cpp | modularity
OutputManager.cpp | tree
PhenotypicTargetHandler.cpp | var_sigma var_tau
Protein.cpp | gen_unit height mean width
ReplicationReport.cpp | indiv
Rna.cpp | gen_unit
StatRecord.cpp
Stats.cpp | exp_m

* Manual fixes
[x] ExpManager.h | t_end
[x] GeneticUnit.cpp | dna rna_list (almost false positive, would have caused no harm)

* Automatic fix all other symbols
: find ! -path '*test*' ! -path './libaevol/SFMT-src-1.4/*' \( -name '*.cpp' -o -name '*.h' \) -print0
: | xargs -0 sed -i -r -e 's/([^a-zA-Z0-9_])_([a-zA-Z0-9][a-zA-Z0-9_]+)/\1\2_/g'

* Minor afterhand fixes
fix OPENMP_ and changed_ with similar `find | xargs sed`s

* Check there are no more _members
: grep -rEo --no-filename '[^a-zA-Z0-9_]_[a-zA-Z0-9][a-zA-Z0-9_]+' --include '*.cpp' --include '*.h' --exclude-dir tests
: | sed -e 's/^.//' | sort -u

Instead of a single class with switches everywhere, mutations are now
handled through a proper polymorphism.

Here is the class hierarchy:
Mutation
  <-- LocalMutation
        <-- PointMutation
        <-- SmallInsertion
        <-- SmallDeletion
  <-- Rearrangement
        <-- Duplication
        <-- Deletion
        <-- Translocation
        <-- Inversion
  <-- HT
        <-- InsertionHT
        <-- ReplacementHT

The call of an individual to its replication report's
signal_end_of_replication is now indirect

The responsability of the creation of the tree was shared between
different entities (the tree of course, but also the individuals
and their dna).
This made it really difficult to deal with.

The solution presented here isn't optimal (yet ?). In particular,
trees had to be temporarily disabled for experiments with plasmids.
The tree now handles itself more independently. It is filled with
naive ReplicationReports which will have to be :
  - initialized at the beginning of a replication
  - filled in at the end of a replication
  - finalized at the end of a generation

In the future, it would be nice to implement that as an Observer
pattern to decouple the involved classes