#!/bin/sh
# Experimental alternative to build2 that can run steps in parallel if their
# dependencies are met. Caveats:
#
#   1. We rely on the OS to schedule steps, so thousands of independent tasks
#      will likely be counter-productive.
#   2. Can run out of virtual memory if you spawn too many say $CC processes.
#   3. Compilation errors can cause the script to hang.
#   4. Ugly as heck! This version really benefits from comparisons with its
#      'upstream', build2. And even then, diff gets confused.
#   5. There's a mechanical difficulty: we use mktemp to reliably create
#      temporary filenames, which has the side effect of also creating the
#      files. So zero-size files are considered equivalent to non-existent
#      files. When some commands have trouble with this (e.g. ar) we need to
#      delete the empty file, which can expose us to a race condition wrt
#      mktemp.

set -e  # stop immediately on error

# Some environment variables that can be passed in. For example, to turn off
# optimization:
#   $ CFLAGS=-g ./build4
test "$CXX" || export CXX=c++
test "$CC" || export CC=cc
test "$CFLAGS" || export CFLAGS="-g -O2"
export CFLAGS="$CFLAGS -Wall -Wextra -ftrapv -fno-strict-aliasing"

# Outline:
# [0-9]*.cc -> mu.cc -> .build/*.cc -> .build/*.o -> .build/mu_bin
# (layers)   |        |              |             |
#          tangle  cleave          $CXX          $CXX

# can also be called with a layer to only build until
#   $ ./build4 --until 050
UNTIL_LAYER=${2:-zzz}

# there's two mechanisms for fast builds here:
# - if a command is quick to run, always run it but update the result only on any change
# - otherwise run it only if the output is 'older_than' the inputs
#
# avoid combining both mechanisms for a single file
# otherwise you'll see spurious messages about files being updated
# risk: a file may unnecessarily update without changes, causing unnecessary work downstream

# return 1 if $1 is older than _any_ of the remaining args
older_than() {
  local target=$1
  shift
  if [ ! -s $target ]
  then
#?     echo "$target has size zero"
#?     echo "updating $target" >&2
    return 0  # success
  fi
  local f
  for f in $*
  do
    if [ $f -nt $target ]
    then
#?       echo "updating $target" >&2
      return 0  # success
    fi
  done
  return 1  # failure
}

# redirect to $1, unless it's already identical
update() {
  if [ ! -e $1 ]
  then
    cat > $1
  else
    cat > $1.tmp
    diff -q $1 $1.tmp >/dev/null  &&  rm $1.tmp  ||  mv $1.tmp $1
  fi
}

update_cp() {
  if [ ! -e $2/$1 ]
  then
    cp $1 $2
  elif [ $1 -nt $2/$1 ]
  then
    cp $1 $2
  fi
}

noisy_cd() {
  cd $1
  echo "-- `pwd`" >&2
}

mv_if_exists() {
  test -e $1  &&  mv $1 $2
  return 0
}

wait_for_all() {
  # could use inotify on Linux
  while ! all_exist "$@"
  do
    sleep 1
  done
}

all_exist() {
  for f in "$@"
  do
    test -e $f  ||  return 1
  done
  return 0
}

TMP=`mktemp`
mv_if_exists enumerate/enumerate $TMP
(
  wait_for_all enumerate/enumerate.cc
  older_than $TMP enumerate/enumerate.cc && {
    echo "building enumerate"
    $CXX $CFLAGS enumerate/enumerate.cc -o $TMP
    echo "done building enumerate"
  }
  mv $TMP enumerate/enumerate
) &

TMP=`mktemp`
mv_if_exists tangle/tangle $TMP
(
  wait_for_all tangle/*.cc
  older_than $TMP tangle/*.cc && {
    echo "building tangle"
    cd tangle
      # auto-generate various lists (ending in '_list' by convention) {
      # list of types
      {
        grep -h "^struct .* {" [0-9]*.cc  |sed 's/\(struct *[^ ]*\).*/\1;/'
        grep -h "^typedef " [0-9]*.cc
      }  |update type_list
      # list of function declarations, so I can define them in any order
      grep -h "^[^ #].*) {" [0-9]*.cc  |sed 's/ {.*/;/'  |update function_list
      # list of code files to compile
      ls [0-9]*.cc  |grep -v "\.test\.cc$"  |sed 's/.*/#include "&"/'  |update file_list
      # list of test files to compile
      ls [0-9]*.test.cc  |sed 's/.*/#include "&"/'  |update test_file_list
      # list of tests to run
      grep -h "^[[:space:]]*void test_" [0-9]*.cc  |sed 's/^\s*void \(.*\)() {$/\1,/'  |update test_list
      # }
      # Now that we have all the _lists, compile 'tangle'
      $CXX $CFLAGS boot.cc -o $TMP
    cd ..
    echo "done building tangle"
  }
  mv $TMP tangle/tangle
) &

wait_for_all enumerate/enumerate
echo "enumerating layers"
LAYERS=$(./enumerate/enumerate --until $UNTIL_LAYER  |grep '\.cc$')
echo "done enumerating layers"

TMP=`mktemp`
mv_if_exists mu.cc $TMP
(
  wait_for_all $LAYERS enumerate/enumerate tangle/tangle
  older_than $TMP $LAYERS enumerate/enumerate tangle/tangle && {
    echo "running tangle"
    # no update here; rely on 'update' calls downstream
    ./tangle/tangle $LAYERS  > $TMP
    echo "done running tangle"
  }
  mv $TMP mu.cc
) &

TMP=`mktemp`
mv_if_exists cleave/cleave $TMP
(
  wait_for_all cleave/cleave.cc
  older_than $TMP cleave/cleave.cc && {
    echo "building cleave"
    $CXX $CFLAGS cleave/cleave.cc -o $TMP
    rm -rf .build
    echo "done building cleave"
  }
  mv $TMP cleave/cleave
) &

wait_for_all mu.cc cleave/cleave  # cleave/cleave just for the .build cleanup
mkdir -p .build
# auto-generate function declarations, so I can define them in any order
# functions start out unindented, have all args on the same line, and end in ') {'
#
#                                      \/ ignore methods
grep -h "^[^[:space:]#].*) {$" mu.cc  |grep -v ":.*("  |sed 's/ {.*/;/'  |update .build/function_list
# auto-generate list of tests to run
grep -h "^\s*void test_" mu.cc  |sed 's/^\s*void \(.*\)() {.*/\1,/'  |update .build/test_list
mkdir -p .build/termbox
update_cp termbox/termbox.h .build/termbox

TMP=`mktemp`
mv_if_exists mu_bin $TMP
(
  wait_for_all mu.cc cleave/cleave termbox/*.c termbox/*.h termbox/*.inl
  older_than $TMP mu.cc *_list cleave/cleave termbox/* && {
    echo "building mu_bin"
    ./cleave/cleave mu.cc .build
    cd .build
      # create the list of global variable declarations from the corresponding definitions
      grep ';' global_definitions_list  |sed 's/[=(].*/;/'  |sed 's/^[^\/# ]/extern &/'  |sed 's/^extern extern /extern /'  |update global_declarations_list
      for f in mu_*.cc
      do
        OBJ=`echo $f |sed 's/\.cc$/.o/'`
        TMP=`mktemp`
        mv_if_exists $OBJ $TMP
        (
          older_than $TMP $f header global_declarations_list function_list test_list && {
            echo "building $OBJ"
            $CXX $CFLAGS -c $f -o $TMP
            echo "done building $OBJ"
          }
          mv $TMP $OBJ
        ) &
      done
    cd ../termbox
      TMP=`mktemp`
      mv_if_exists utf8.o $TMP
      (
        older_than $TMP utf8.c && {
          echo "building termbox/utf8.o"
          $CC $CFLAGS -c utf8.c -o $TMP
          echo "done building termbox/utf8.o"
        }
        mv $TMP utf8.o
      ) &
      TMP=`mktemp`
      mv_if_exists termbox.o $TMP
      (
        older_than $TMP termbox.c termbox.h input.inl output.inl bytebuffer.inl && {
          echo "building termbox/termbox.o"
          $CC $CFLAGS -c termbox.c -o $TMP
          echo "done building termbox/termbox.o"
        }
        mv $TMP termbox.o
      ) &
      TMP=`mktemp`
      mv_if_exists libtermbox.a $TMP
      (
        wait_for_all termbox.o utf8.o
        older_than $TMP termbox.o utf8.o && {
          echo "building termbox/libtermbox.a"
          rm $TMP;  ar rcs $TMP termbox.o utf8.o  # race condition; later mktemp may end up reusing this file
          echo "done building termbox/libtermbox.a"
        }
        mv $TMP libtermbox.a
      ) &
    cd ..
    MU_OBJS=`echo .build/mu_*.cc |sed 's/\.cc/.o/g'`
    echo wait_for_all $MU_OBJS termbox/libtermbox.a
    wait_for_all $MU_OBJS termbox/libtermbox.a
    echo "building .build/mu_bin"
    $CXX $CFLAGS $MU_OBJS termbox/libtermbox.a -o $TMP
    echo "done building .build/mu_bin"
    echo "done building mu_bin"
  }
  mv $TMP mu_bin
) &

## [0-9]*.mu -> core.mu

wait_for_all enumerate/enumerate
echo "building core.mu"
MU_LAYERS=$(./enumerate/enumerate --until $UNTIL_LAYER  |grep '\.mu$') || exit 0  # ok if no .mu files
cat $MU_LAYERS  |update core.mu
echo "done building core.mu"

wait_for_all mu_bin
exit 0

# scenarios considered:
#   0 status when nothing needs updating
#   no output when nothing needs updating
#     no output for mu.cc when .mu files modified
#     touch mu.cc but don't modify it; no output on second build
#     touch a .cc layer but don't modify it; no output on second build
#   only a single layer is recompiled when changing a C++ function
#   stop immediately after failure in tangle