Translate Prodigal initialization functions into sparse code to reduc…

…e library size

setup.py

@@ -1,6 +1,8 @@
+import collections
 import configparser
 import functools
 import glob
+import itertools
 import multiprocessing.pool
 import os
 import platform
@@ -520,6 +522,23 @@ class build_clib(_build_clib):
  """A custom `build_clib` that splits the `training.c` file from Prodigal.
  """
 
+ _TrainingInfo = collections.namedtuple(
+ "_TrainingInfo",
+ [
+ "gc",
+ "tt",
+ "st_wt",
+ "bias",
+ "type_wt",
+ "uses_sd",
+ "rbs_wt",
+ "ups_comp",
+ "mot_wt",
+ "no_mot",
+ "gene_dc"
+ ]
+ )
+
  # --- Compatibility with `setuptools.Command`
 
  user_options = _build_clib.user_options + [
@@ -610,25 +629,80 @@ def get_library(self, name):
 
  # --- Build code ---
 
- def _write_source_split(self, training_temp, index, lines):
- filename = os.path.join(training_temp, "training{:02}.c".format(index))
+ def _write_header(self, training_temp, index, lines):
+ filename = os.path.join(training_temp, "training_header.c".format(index))
  with open(filename, "wb") as dst:
- if index != 0:
- dst.write(b'#include "training.h"\n')
  dst.writelines(lines)
  lines.clear()
 
+ def _write_source_split(self, training_temp, index, lines):
+ filename = os.path.join(training_temp, "training{:02}.c".format(index))
+
+ tinf = self._TrainingInfo(*eval(
+ b"[" 
+ + b''.join(lines[2:-4])
+ .replace(b"{", b"[")
+ .replace(b"}", b"]")
+ .rstrip(b";")
+ .strip() 
+ + b"]"
+ ))
+ lines.clear()
+
+ counts = collections.Counter()
+ for i,j,k in itertools.product(range(4), range(4), range(4096)):
+ counts[ tinf.mot_wt[i][j][k] ] += 1
+ background = counts.most_common(1)[0][0]
+
+ with open(filename, "w") as dst:
+ dst.write('#include "training.h"\n')
+
+ dst.write(f"void initialize_metagenome_{index}(struct _training *tptr) {{\n")
+ dst.write(f" tptr->gc = {tinf.gc};\n")
+ dst.write(f" tptr->trans_table = {tinf.tt};\n")
+ dst.write(f" tptr->st_wt = {tinf.st_wt};")
+ for i,b in enumerate(tinf.bias):
+ dst.write(f" tptr->bias[{i}] = {b};\n")
+ for i,w in enumerate(tinf.type_wt):
+ dst.write(f" tptr->type_wt[{i}] = {w};\n")
+ dst.write(f" tptr->uses_sd = {tinf.uses_sd};\n")
+
+ dst.write(f" double rbs_wt[28] = {{ {', '.join(map(str, tinf.rbs_wt))} }};\n")
+ dst.write(f" memcpy(tptr->rbs_wt, rbs_wt, 28 * sizeof(double));\n")
+
+ for i,x in enumerate(tinf.ups_comp):
+ for j,y in enumerate(x):
+ dst.write(f" tptr->ups_comp[{i}][{j}] = {y};\n")
+
+ dst.write(f" tptr->no_mot = {tinf.no_mot};\n")
+ dst.write(f" double gene_dc[4096] = {{ {', '.join(map(str, tinf.gene_dc))} }};\n")
+ dst.write(f" memcpy(tptr->gene_dc, gene_dc, 4096 * sizeof(double));\n")
+
+ dst.write(f" for (size_t i = 0; i < 4; i++)\n")
+ dst.write(f" for (size_t j = 0; j < 4; j++)\n")
+ dst.write(f" for (size_t k = 0; k < 4096; k++)\n")
+ dst.write(f" tptr->mot_wt[i][j][k] = {background};\n")
+
+ for i,j,k in itertools.product(range(4), range(4), range(4096)):
+ wt = tinf.mot_wt[i][j][k]
+ if wt != background:
+ dst.write(f" tptr->mot_wt[{i}][{j}][{k}] = {wt};\n")
+ dst.write("}\n")
+
  def _split_training_source(self, training_file, training_temp):
  self.mkpath(training_temp)
  with open(training_file, "rb") as src:
  lines = []
  index = 0
  for line in src:
  if line.startswith(b"void initialize_metagenome"):
- self._write_source_split(training_temp, index, lines)
- index += 1
- if line.lstrip().startswith(b"struct _training"):
- line = line.replace(b"struct _training", b"static const struct _training")
+ if line.startswith(b"void initialize_metagenome_0"):
+ self._write_header(training_temp, index, lines)
+ else:
+ self._write_source_split(training_temp, index, lines)
+ index += 1
+ # if line.lstrip().startswith(b"struct _training"):
+ # line = line.replace(b"struct _training", b"static const struct _training")
  lines.append(line)
  self._write_source_split(training_temp, index, lines)
 
@@ -705,6 +779,7 @@ def build_library(self, library):
  if library.name == "prodigal":
  sources.remove(self.training_file)
  sources.extend(sorted(glob.iglob(os.path.join(self.training_temp, "*.c"))))
+ # sources.append("build/src/training.c")
  objects = [
  os.path.join(self.build_temp, s.replace(".c", self.compiler.obj_extension))
  for s in sources