Skip to content
Find file
Fetching contributors…
Cannot retrieve contributors at this time
executable file 1120 lines (959 sloc) 49 KB
#!/usr/bin/env python
emcc - compiler helper script
emcc is a drop-in replacement for a compiler like gcc or clang.
Tell your build system to use this instead of the compiler, and similarly
use emar, emranlib etc. instead of the same command without 'em'.
Example uses:
* For configure, instead of ./configure, cmake, etc., run
with that command as an argument, for example ./configure [options] is a tiny script that just sets some environment vars
as a convenience. The command just shown is equivalent to
EMMAKEN_JUST_CONFIGURE=1 RANLIB=PATH/emranlib AR=PATH/emar CXX=PATH/em++ CC=PATH/emcc ./configure [options]
where PATH is the path to this file.
EMMAKEN_JUST_CONFIGURE tells emcc that it is being run in ./configure,
so it should relay everything to gcc/g++. You should not define that when
running make, of course.
* With CMake, the same command will work (with cmake instead of ./configure). You may also be
able to do the following in your CMakeLists.txt:
* For SCons the can be imported like so:
__file__ = str(Dir('#/project_path_to_emscripten/dummy/dummy'))
__rootpath__ = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
def path_from_root(*pathelems):
return os.path.join(__rootpath__, *pathelems)
sys.path += [path_from_root('')]
from tools.shared import *
For using the Emscripten compilers/linkers/etc. you can do:
env = Environment()
env.Replace(LINK = LLVM_LD)
env.Replace(LD = LLVM_LD)
TODO: Document all relevant setup changes
After setting that up, run your build system normally.
Note the appearance of em++ instead of emcc
for the C++ compiler. This is needed for cases where we get
a C++ file with a C extension, in which case CMake can be told
to run g++ on it despite the .c extension, see
(If a similar situation occurs with ./configure, you can do the same there too.)
emcc can be influenced by a few environment variables:
EMMAKEN_NO_SDK - Will tell emcc *not* to use the emscripten headers. Instead
your system headers will be used.
EMMAKEN_COMPILER - The compiler to be used, if you don't want the default clang.
import os, sys, shutil, tempfile, subprocess, shlex
from subprocess import PIPE, STDOUT
from tools import shared
from tools.shared import Compression, execute, suffix, unsuffixed, unsuffixed_basename
# Mapping of emcc opt levels to llvm opt levels. We use llvm opt level 3 in emcc opt
# levels 2 and 3 (emcc 3 is unsafe opts, so unsuitable for the only level to get
# llvm opt level 3, and speed-wise emcc level 2 is already the slowest/most optimizing
# level)
0: 0,
1: 1,
2: 3,
3: 3,
DEBUG = os.environ.get('EMCC_DEBUG')
TEMP_DIR = os.environ.get('EMCC_TEMP_DIR')
LEAVE_INPUTS_RAW = os.environ.get('EMCC_LEAVE_INPUTS_RAW') # Do not compile .ll files into .bc, just compile them with emscripten directly
# Not recommended, this is mainly for the test runner, or if you have some other
# specific need.
# One major limitation with this mode is that dlmalloc and libc++ cannot be
# added in. Also, LLVM optimizations will not be done, nor dead code elimination
AUTODEBUG = os.environ.get('EMCC_AUTODEBUG') # If set to 1, we will run the autodebugger (the automatic debugging tool, see tools/autodebugger).
# Note that this will disable inclusion of libraries. This is useful because including
# dlmalloc makes it hard to compare native and js builds
EMCC_CFLAGS = os.environ.get('EMCC_CFLAGS') # Additional compiler flags that we treat as if they were passed to us on the commandline
if DEBUG: print >> sys.stderr, '\nemcc invocation: ', ' '.join(sys.argv), (' + ' + EMCC_CFLAGS if EMCC_CFLAGS else '')
if EMCC_CFLAGS: sys.argv.append(EMCC_CFLAGS)
if DEBUG and LEAVE_INPUTS_RAW: print >> sys.stderr, 'emcc: leaving inputs raw'
stdout = PIPE if not DEBUG else None # suppress output of child processes
stderr = PIPE if not DEBUG else None # unless we are in DEBUG mode
# Handle some global flags
if len(sys.argv) == 1:
print 'emcc: no input files'
if sys.argv[1] == '--version':
print '''emcc (Emscripten GCC-like replacement) 2.0
Copyright (C) 2012 the Emscripten authors.
This is free and open source software under the MIT license.
elif sys.argv[1] == '--help':
this = os.path.basename('em++' if os.environ.get('EMMAKEN_CXX') else 'emcc')
print '''%s [options] file...
Most normal gcc/g++ options will work, for example:
--help Display this information
--version Display compiler version information
Options that are modified or new in %s include:
-O0 No optimizations (default)
-O1 Simple optimizations, including LLVM -O1
optimizations, and no runtime assertions
or C++ exception catching (to re-enable
C++ exception catching, use
(For details on the affects of different
opt levels, see apply_opt_level() in
tools/ and also src/settings.js.)
Note: Optimizations are only done when
compiling to JavaScript, not to intermediate
-O2 As -O1, plus the relooper (loop recreation),
plus closure compiler advanced opts, plus
LLVM -O2 optimizations
Warning: Compiling with this takes a long time!
-O3 As -O2, plus dangerous optimizations that may
break the generated code! If that happens, try
-O2 and then adding dangerous optimizations one
by one.
-s OPTION=VALUE JavaScript code generation option passed
into the emscripten compiler. For the
available options, see src/settings.js
Note that for options that are lists, you
need quotation marks in most shells, for
(without the external "s in either of those,
you would get an error)
--typed-arrays <mode> 0: No typed arrays
1: Parallel typed arrays
2: Shared (C-like) typed arrays (default)
--llvm-opts <level> 0: No LLVM optimizations (default in -O0)
1: -O1 LLVM optimizations (default in -O1)
2: -O2 LLVM optimizations
3: -O3 LLVM optimizations (default in -O2+)
--llvm-lto <level> 0: No LLVM LTO (default in -O0)
1: LLVM LTO (default in -O1+)
Note: If LLVM optimizations are not run
(see --llvm-opts), setting this to 1 has no
--closure <on> 0: No closure compiler (default in -O0, -O1)
1: Run closure compiler (default in -O2, -O3)
--js-transform <cmd> <cmd> will be called on the generated code
before it is optimized. This lets you modify
the JavaScript, for example adding some code
or removing some code, in a way that those
modifications will be optimized together with
the generated code properly. <cmd> will be
called with the filename of the generated
code as a parameter; to modify the code, you
can read the original data and then append to
it or overwrite it with the modified data.
<cmd> is interpreted as a space-separated
list of arguments, for example, <cmd> of
"python" will cause a python
script to be run.
--pre-js <file> A file whose contents are added before the
generated code. This is done *before*
optimization, so it will be minified
properly if closure compiler is run.
--post-js <file> A file whose contents are added after the
generated code This is done *before*
optimization, so it will be minified
properly if closure compiler is run.
--embed-file <file> A file to embed inside the generated
JavaScript. The compiled code will be able
to access the file in the current directory
with the same name as given here. So if
you do --embed-file dir/file.dat, then
(1) dir/file.dat must exist relative to
where you run emcc, and (2) your compiled
code will be able to find the file by
reading that same path, dir/file.dat.
If a directory is passed here, its entire
contents will be embedded.
--preload-file <name> A file to preload before running the
compiled code asynchronously. Otherwise
similar to --embed-file, except that this
option is only relevant when generating
HTML (it uses asynchronous binary XHRs),
or JS that will be used in a web page.
If a directory is passed here, its entire
contents will be preloaded.
Preloaded files are stored in,
where filename.html is the main file you
are compiling to. To run your code, you
will need both the .html and the .data.
--compression <codec> Compress both the compiled code and embedded/
preloaded files. <codec> should be a triple,
where native_encoder is a native executable
that compresses stdin to stdout (the simplest
possible interface), js_decoder is a
JavaScript file that implements a decoder,
and js_name is the name of the function to
call in the decoder file (which should
receive an array/typed array and return
an array/typed array.
Compression only works when generating HTML.
When compression is on, all filed specified
to be preloaded are compressed in one big
archive, which is given the same name as the
output HTML but with suffix .data.compress
--minify <on> 0: Do not minify the generated JavaScript's
whitespace (default if closure compiler
will not be run)
1: Minify the generated JavaScript's
whitespace (default if closure compiler
will be run). Note that this by itself
will not minify the code (closure does
--split <size> Splits the resulting javascript file into pieces
to ease debugging. This option only works if
Javascript is generated (target -o <name>.js).
Files with function declarations must be loaded
before main file upon execution.
Without "-g" option:
Creates files with function declarations up
to the given size with the suffix
"_functions.partxxx.js" and a main file with
the suffix ".js".
With "-g" option:
Recreates the directory structure of the C
source files and stores function declarations
in their respective C files with the suffix
".js". If such a file exceeds the given size,
files with the suffix ".partxxx.js" are created.
The main file resides in the base directory and
has the suffix ".js".
--ignore-dynamic-linking Normally emcc will treat dynamic linking like
static linking, by linking in the code from
the dynamic library. This fails if the same
dynamic library is linked more than once.
With this option, dynamic linking is ignored,
which allows the build system to proceed without
errors. However, you will need to manually
link to the shared libraries later on yourself.
--shell-file <path> The path name to a skeleton HTML file used
when generating HTML output. The shell file
used needs to have this token inside it:
Note that this argument is ignored if a
target other than HTML is specified using
the -o option.
--js-library <lib> A JavaScript library to use in addition to
those in Emscripten's src/library_*
-v Turns on verbose output. This will pass
-v to Clang, and also enable EMCC_DEBUG
to details emcc's operations
--remove-duplicates If set, will remove duplicate symbols when
linking. This can be useful because
llvm-link's behavior is not as permissive
as ld is.
--clear-cache Manually clears the cache of compiled
emscripten system libraries (libc++,
libc++abi, dlmalloc). This is normally
handled automatically, but if you update
llvm in-place (instead of having a different
directory for a new version), the caching
mechanism can get confused. Clearing the
cache can fix weird problems related to
cache incompatibilities, like clang failing
to link with library files.
The target file, if specified (-o <target>), defines what will
be generated:
<name>.js JavaScript (default)
<name>.html HTML with embedded JavaScript
<name>.bc LLVM bitcode
<name>.o LLVM bitcode (same as .bc)
The -c option (which tells gcc not to run the linker) will
cause LLVM bitcode to be generated, as %s only generates
JavaScript in the final linking stage of building.
The input file(s) can be either source code files that
Clang can handle (C or C++), LLVM bitcode in binary form,
or LLVM assembly files in human-readable form.
emcc is affected by several environment variables. For details, view
the source of emcc (search for 'os.environ').
''' % (this, this, this)
elif len(sys.argv) == 2 and sys.argv[1] == '-v': # -v with no inputs
print 'emcc (Emscripten GCC-like replacement) 2.0'
exit([shared.CLANG, '-v']))
def is_minus_s_for_emcc(newargs,i):
assert newargs[i] == '-s'
if i+1 < len(newargs) and '=' in newargs[i+1]: # -s OPT=VALUE is for us, -s by itself is a linker option
return True
print >> sys.stderr, 'emcc: warning: treating -s as linker option and not as -s OPT=VALUE for js compilation'
return False
# If this is a configure-type thing, do not compile to JavaScript, instead use clang
# to compile to a native binary (using our headers, so things make sense later)
CONFIGURE_CONFIG = os.environ.get('EMMAKEN_JUST_CONFIGURE') or 'conftest.c' in sys.argv
CMAKE_CONFIG = 'CMakeFiles/cmTryCompileExec.dir' in ' '.join(sys.argv)# or 'CMakeCCompilerId' in ' '.join(sys.argv)
compiler = shared.CLANG
if not ('CXXCompiler' in ' '.join(sys.argv) or os.environ.get('EMMAKEN_CXX')):
compiler = shared.to_cc(compiler)
def filter_emscripten_options(argv):
idx = 0
skip_next = False
for el in argv:
if skip_next:
skip_next = False
idx += 1
if el == '-s' and is_minus_s_for_emcc(argv, idx):
skip_next = True
yield el
idx += 1
cmd = [compiler] + shared.EMSDK_OPTS + ['-DEMSCRIPTEN'] + list(filter_emscripten_options(sys.argv[1:]))
if DEBUG: print >> sys.stderr, 'emcc, just configuring: ', ' '.join(cmd)
if os.environ.get('EMMAKEN_COMPILER'):
CXX = os.environ['EMMAKEN_COMPILER']
CXX = shared.CLANG
CC = shared.to_cc(CXX)
# If we got here from a redirection through, then force a C++ compiler here
if os.environ.get('EMMAKEN_CXX'):
# ---------------- Utilities ---------------
SOURCE_SUFFIXES = ('.c', '.cpp', '.cxx', '.cc')
BITCODE_SUFFIXES = ('.bc', '.o', '.obj')
DYNAMICLIB_SUFFIXES = ('.dylib', '.so', '.dll')
LIB_PREFIXES = ('', 'lib')
seen_names = {}
def uniquename(name):
if name not in seen_names:
seen_names[name] = str(len(seen_names))
return unsuffixed(name) + '_' + seen_names[name] + (('.' + suffix(name)) if suffix(name) else '')
# ---------------- End configs -------------
if len(sys.argv) == 1 or sys.argv[1] in ['x', 't']:
# noop ar
if DEBUG: print >> sys.stderr, 'emcc, just ar'
use_cxx = True
header = False # pre-compiled headers. We fake that by just copying the file
for i in range(1, len(sys.argv)):
arg = sys.argv[i]
if not arg.startswith('-'):
if arg.endswith('.c'):
use_cxx = False
if arg.endswith('.h') and sys.argv[i-1] != '-include':
header = True
if '-M' in sys.argv or '-MM' in sys.argv:
# Just output dependencies, do not compile. Warning: clang and gcc behave differently with -MF! (clang seems to not recognize it)
cmd = [CC] + shared.COMPILER_OPTS + sys.argv[1:]
if DEBUG: print >> sys.stderr, 'emcc, just dependencies: ', ' '.join(cmd)
# Check if a target is specified
target = None
for i in range(len(sys.argv)-1):
if sys.argv[i].startswith('-o='):
raise Exception('Invalid syntax: do not use -o=X, use -o X')
if sys.argv[i] == '-o':
target = sys.argv[i+1]
sys.argv = sys.argv[:i] + sys.argv[i+2:]
if header: # header or such
if len(sys.argv) >= 3: # if there is a source and a target, then copy, otherwise do nothing
sys.argv = filter(lambda arg: not arg.startswith('-I'), sys.argv)
if DEBUG: print >> sys.stderr, 'Just copy:', sys.argv[-1], target
shutil.copy(sys.argv[-1], target)
if DEBUG: print >> sys.stderr, 'No-op.'
temp_dir = TEMP_DIR
if os.path.exists(temp_dir):
shutil.rmtree(temp_dir) # clear it
temp_root = os.path.join(shared.TEMP_DIR, 'emcc')
if not os.path.exists(temp_root):
temp_dir = tempfile.mkdtemp(dir=temp_root)
def in_temp(name):
return os.path.join(temp_dir, os.path.basename(name))
call = CXX if use_cxx else CC
## Parse args
newargs = sys.argv[1:]
opt_level = 0
llvm_opts = None
llvm_lto = None
closure = None
js_transform = None
pre_js = ''
post_js = ''
minify_whitespace = None
split_js_file = None
preload_files = []
embed_files = []
compression = None
ignore_dynamic_linking = False
shell_path = shared.path_from_root('src', 'shell.html')
js_libraries = []
remove_duplicates = False
def check_bad_eq(arg):
assert '=' not in arg, 'Invalid parameter (do not use "=" with "--" options)'
absolute_warning_shown = False
for i in range(len(newargs)):
if newargs[i].startswith('-O'):
opt_level = int(newargs[i][2])
assert 0 <= opt_level <= 3
raise Exception('Invalid optimization level: ' + newargs[i])
newargs[i] = ''
elif newargs[i].startswith('--llvm-opts'):
llvm_opts = eval(newargs[i+1])
newargs[i] = ''
newargs[i+1] = ''
elif newargs[i].startswith('--llvm-lto'):
llvm_lto = eval(newargs[i+1])
newargs[i] = ''
newargs[i+1] = ''
elif newargs[i].startswith('--closure'):
closure = int(newargs[i+1])
newargs[i] = ''
newargs[i+1] = ''
elif newargs[i].startswith('--js-transform'):
js_transform = newargs[i+1]
newargs[i] = ''
newargs[i+1] = ''
elif newargs[i].startswith('--pre-js'):
pre_js += open(newargs[i+1]).read() + '\n'
newargs[i] = ''
newargs[i+1] = ''
elif newargs[i].startswith('--post-js'):
post_js += open(newargs[i+1]).read() + '\n'
newargs[i] = ''
newargs[i+1] = ''
elif newargs[i].startswith('--minify'):
minify_whitespace = int(newargs[i+1])
newargs[i] = ''
newargs[i+1] = ''
elif newargs[i].startswith('--split'):
split_js_file = int(newargs[i+1])
newargs[i] = ''
newargs[i+1] = ''
elif newargs[i].startswith('--embed-file'):
newargs[i] = ''
newargs[i+1] = ''
elif newargs[i].startswith('--preload-file'):
newargs[i] = ''
newargs[i+1] = ''
elif newargs[i].startswith('--compression'):
parts = newargs[i+1].split(',')
assert len(parts) == 3, '--compression requires specifying native_encoder,js_decoder,js_name - see emcc --help. got: %s' % newargs[i+1]
Compression.encoder = parts[0]
Compression.decoder = parts[1]
Compression.js_name = parts[2]
assert os.path.exists(Compression.encoder), 'native encoder %s does not exist' % Compression.encoder
assert os.path.exists(Compression.decoder), 'js decoder %s does not exist' % Compression.decoder
Compression.on = True
newargs[i] = ''
newargs[i+1] = ''
elif newargs[i] == '--ignore-dynamic-linking':
ignore_dynamic_linking = True
newargs[i] = ''
elif newargs[i] == '-v':
shared.COMPILER_OPTS += ['-v']
newargs[i] = ''
elif newargs[i].startswith('--shell-file'):
shell_path = newargs[i+1]
newargs[i] = ''
newargs[i+1] = ''
elif newargs[i].startswith('--js-library'):
newargs[i] = ''
newargs[i+1] = ''
elif newargs[i] == '--remove-duplicates':
remove_duplicates = True
newargs[i] = ''
elif newargs[i] == '--clear-cache':
newargs[i] = ''
print >> sys.stderr, 'emcc: clearing cache'
elif newargs[i].startswith(('-I/', '-L/')):
if not absolute_warning_shown:
print >> sys.stderr, 'emcc: warning: -I or -L of an absolute path encountered. If this is to a local system header/library, it may cause problems (local system files make sense for compiling natively on your system, but not necessarily to JavaScript)' # Of course an absolute path to a non-system-specific library or header is fine, and you can ignore this warning. The danger are system headers that are e.g. x86 specific and nonportable. The emscripten bundled headers are modified to be portable, local system ones are generally not
absolute_warning_shown = True
newargs = [ arg for arg in newargs if arg is not '' ]
if llvm_opts is None: llvm_opts = LLVM_OPT_LEVEL[opt_level]
if llvm_lto is None: llvm_lto = llvm_opts > 0
if closure is None: closure = 1 if opt_level >= 2 else 0
if minify_whitespace is None:
minify_whitespace = closure # if closure is run, minify whitespace
if closure:
assert os.path.exists(shared.CLOSURE_COMPILER), 'emcc: fatal: Closure compiler (%s) does not exist' % shared.CLOSURE_COMPILER
settings_changes = []
for i in range(len(newargs)):
if newargs[i] == '-s':
if is_minus_s_for_emcc(newargs, i):
newargs[i] = newargs[i+1] = ''
elif newargs[i].startswith('--typed-arrays'):
assert '=' not in newargs[i], 'Invalid typed arrays parameter (do not use "=")'
settings_changes.append('USE_TYPED_ARRAYS=' + newargs[i+1])
newargs[i] = ''
newargs[i+1] = ''
newargs = [ arg for arg in newargs if arg is not '' ]
if split_js_file:
# Find input files
input_files = []
has_source_inputs = False
lib_dirs = [shared.path_from_root('system', 'local', 'lib'),
shared.path_from_root('system', 'lib')]
libs = []
for i in range(len(newargs)): # find input files XXX this a simple heuristic. we should really analyze based on a full understanding of gcc params,
# right now we just assume that what is left contains no more |-x OPT| things
arg = newargs[i]
if i > 0:
prev = newargs[i-1]
if prev in ['-MT', '-install_name']: continue # ignore this gcc-style argument
if arg.endswith(SOURCE_SUFFIXES + BITCODE_SUFFIXES + DYNAMICLIB_SUFFIXES + ASSEMBLY_SUFFIXES) or shared.Building.is_ar(arg): # we already removed -o <target>, so all these should be inputs
newargs[i] = ''
if os.path.exists(arg):
if arg.endswith(SOURCE_SUFFIXES):
has_source_inputs = True
# this should be bitcode, make sure it is valid
if arg.endswith(ASSEMBLY_SUFFIXES) or shared.Building.is_bitcode(arg):
# if it's not, and it's a library, just add it to libs to find later
l = unsuffixed_basename(arg)
for prefix in LIB_PREFIXES:
if not prefix: continue
if l.startswith(prefix):
l = l[len(prefix):]
newargs[i] = ''
print >> sys.stderr, 'emcc: %s: warning: Not valid LLVM bitcode' % arg
print >> sys.stderr, 'emcc: %s: warning: No such file or directory' % arg
elif arg.startswith('-L'):
newargs[i] = ''
elif arg.startswith('-l'):
newargs[i] = ''
original_input_files = input_files[:]
newargs = [ arg for arg in newargs if arg is not '' ]
# Find library files
for lib in libs:
if DEBUG: print >> sys.stderr, 'emcc: looking for library "%s"' % lib
found = False
for prefix in LIB_PREFIXES:
name = prefix + lib + suff
for lib_dir in lib_dirs:
path = os.path.join(lib_dir, name)
if os.path.exists(path):
if DEBUG: print >> sys.stderr, 'emcc: found library "%s" at %s' % (lib, path)
found = True
if found: break
if found: break
if ignore_dynamic_linking:
input_files = filter(lambda input_file: not input_file.endswith(DYNAMICLIB_SUFFIXES), input_files)
if len(input_files) == 0:
print >> sys.stderr, 'emcc: no input files'
print >> sys.stderr, 'note that input files without a known suffix are ignored, make sure your input files end with one of: ' + str(SOURCE_SUFFIXES + BITCODE_SUFFIXES + DYNAMICLIB_SUFFIXES + STATICLIB_SUFFIXES + ASSEMBLY_SUFFIXES)
specified_target = target
target = specified_target if specified_target is not None else 'a.out.js' # specified_target is the user-specified one, target is what we will generate
target_basename = unsuffixed_basename(target)
# -c means do not link in gcc, and for us, the parallel is to not go all the way to JS, but stop at bitcode
has_dash_c = '-c' in newargs
if has_dash_c:
assert has_source_inputs, 'Must have source code inputs to use -c'
target = target_basename + '.o'
if '.' in target:
final_suffix = target.split('.')[-1]
final_suffix = ''
assert not (Compression.on and final_suffix != 'html'), 'Compression only works when generating HTML'
# Apply optimization level settings
shared.Settings.apply_opt_level(opt_level, noisy=True)
# Apply -s settings in newargs here (after optimization levels, so they can override them)
for change in settings_changes:
key, value = change.split('=')
exec('shared.Settings.' + key + ' = ' + value)
## Compile source code to bitcode
if DEBUG: print >> sys.stderr, 'emcc: compiling to bitcode'
temp_files = []
# First, generate LLVM bitcode. For each input file, we get base.o with bitcode
for input_file in input_files:
if input_file.endswith(SOURCE_SUFFIXES):
if DEBUG: print >> sys.stderr, 'emcc: compiling source file: ', input_file
output_file = in_temp(unsuffixed(uniquename(input_file)) + '.o')
args = newargs + ['-emit-llvm', '-c', input_file, '-o', output_file]
if DEBUG: print >> sys.stderr, "emcc running:", call, ' '.join(args)
execute([call] + args) # let compiler frontend print directly, so colors are saved (PIPE kills that)
if not os.path.exists(output_file):
print >> sys.stderr, 'emcc: compiler frontend failed to generate LLVM bitcode, halting'
else: # bitcode
if input_file.endswith(BITCODE_SUFFIXES):
if DEBUG: print >> sys.stderr, 'emcc: copying bitcode file: ', input_file
temp_file = in_temp(unsuffixed(uniquename(input_file)) + '.o')
shutil.copyfile(input_file, temp_file)
elif input_file.endswith(DYNAMICLIB_SUFFIXES) or shared.Building.is_ar(input_file):
if DEBUG: print >> sys.stderr, 'emcc: copying library file: ', input_file
temp_file = in_temp(uniquename(input_file))
shutil.copyfile(input_file, temp_file)
else: #.ll
# Note that by assembling the .ll file, then disassembling it later, we will
# remove annotations which is a good thing for compilation time
if DEBUG: print >> sys.stderr, 'emcc: assembling assembly file: ', input_file
temp_file = in_temp(unsuffixed(uniquename(input_file)) + '.o')
shared.Building.llvm_as(input_file, temp_file)
if not LEAVE_INPUTS_RAW: assert len(temp_files) == len(input_files)
# If we were just asked to generate bitcode, stop there
if final_suffix not in ['js', 'html']:
if llvm_opts > 0:
print >> sys.stderr, 'emcc: warning: -Ox flags ignored, since not generating JavaScript'
if not specified_target:
for input_file in input_files:
shutil.move(in_temp(unsuffixed(uniquename(input_file)) + '.o'), unsuffixed_basename(input_file) + '.' + final_suffix)
if len(input_files) == 1:
shutil.move(in_temp(unsuffixed(uniquename(input_files[0])) + '.o'), specified_target)
assert len(original_input_files) == 1 or not has_dash_c, 'fatal error: cannot specify -o with -c with multiple files' + str(sys.argv) + ':' + str(original_input_files)
# We have a specified target (-o <target>), which is not JavaScript or HTML, and
# we have multiple files: Link them
if DEBUG: print >> sys.stderr, 'emcc: link: ' + str(temp_files), specified_target, remove_duplicates=remove_duplicates)
## Continue on to create JavaScript
if DEBUG: print >> sys.stderr, 'emcc: will generate JavaScript'
extra_files_to_link = []
# Check if we need to include some libraries that we compile. (We implement libc ourselves in js, but
# compile a malloc implementation and stdlibc++.)
# Note that we assume a single symbol is enough to know if we have/do not have dlmalloc etc. If you
# include just a few symbols but want the rest, this will not work.
# dlmalloc
def create_dlmalloc():
if DEBUG: print >> sys.stderr, 'emcc: building dlmalloc for cache'
execute(shared.ENV_PREFIX + ['python', shared.EMCC, shared.path_from_root('system', 'lib', 'dlmalloc.c'), '-g', '-o', in_temp('dlmalloc.o')], stdout=stdout, stderr=stderr)
# we include the libc++ new stuff here, so that the common case of using just new/delete is quick to link
execute(shared.ENV_PREFIX + ['python', shared.EMXX, shared.path_from_root('system', 'lib', 'libcxx', 'new.cpp'), '-g', '-o', in_temp('new.o')], stdout=stdout, stderr=stderr)[in_temp('dlmalloc.o'), in_temp('new.o')], in_temp('dlmalloc_full.o'))
return in_temp('dlmalloc_full.o')
def fix_dlmalloc():
# dlmalloc needs some sign correction. # If we are in mode 0, switch to 2. We will add our lines
if shared.Settings.CORRECT_SIGNS == 0: raise Exception('we need to change to 2')
except: # we fail if equal to 0 - so we need to switch to 2 - or if CORRECT_SIGNS is not even in Settings
shared.Settings.CORRECT_SIGNS = 2
if shared.Settings.CORRECT_SIGNS == 2:
shared.Settings.CORRECT_SIGNS_LINES = [shared.path_from_root('src', 'dlmalloc.c') + ':' + str(i+4) for i in [4816, 4191, 4246, 4199, 4205, 4235, 4227]]
# If we are in mode 1, we are correcting everything anyhow. If we are in mode 3, we will be corrected
# so all is well anyhow too.
# XXX We also need to add libc symbols that use malloc, for example strdup. It's very rare to use just them and not
# a normal malloc symbol (like free, after calling strdup), so we haven't hit this yet, but it is possible.
dlmalloc_symbols = open(shared.path_from_root('system', 'lib', 'dlmalloc.symbols')).read().split('\n')
# libcxx
def create_libcxx():
if DEBUG: print >> sys.stderr, 'emcc: building libcxx for cache'
shared.Building.build_library('libcxx', shared.EMSCRIPTEN_TEMP_DIR, shared.EMSCRIPTEN_TEMP_DIR, ['libcxx.bc'], configure=None, copy_project=True, source_dir=shared.path_from_root('system', 'lib', 'libcxx'))
return os.path.join(shared.EMSCRIPTEN_TEMP_DIR, 'libcxx', 'libcxx.bc')
def fix_libcxx():
assert shared.Settings.QUANTUM_SIZE == 4, 'We do not support libc++ with QUANTUM_SIZE == 1'
# libcxx might need corrections, so turn them all on. TODO: check which are actually needed
shared.Settings.CORRECT_SIGNS = shared.Settings.CORRECT_OVERFLOWS = shared.Settings.CORRECT_ROUNDINGS = 1
#print >> sys.stderr, 'emcc: info: using libcxx turns on CORRECT_* options'
libcxx_symbols = map(lambda line: line.strip().split(' ')[1], open(shared.path_from_root('system', 'lib', 'libcxx', 'symbols')).readlines())
libcxx_symbols = filter(lambda symbol: symbol not in dlmalloc_symbols, libcxx_symbols)
libcxx_symbols = set(libcxx_symbols)
# libcxxabi - just for dynamic_cast for now
def create_libcxxabi():
if DEBUG: print >> sys.stderr, 'emcc: building libcxxabi for cache'
shared.Building.build_library('libcxxabi', shared.EMSCRIPTEN_TEMP_DIR, shared.EMSCRIPTEN_TEMP_DIR, ['libcxxabi.bc'], configure=None, copy_project=True, source_dir=shared.path_from_root('system', 'lib', 'libcxxabi'))
return os.path.join(shared.EMSCRIPTEN_TEMP_DIR, 'libcxxabi', 'libcxxabi.bc')
def fix_libcxxabi():
assert shared.Settings.QUANTUM_SIZE == 4, 'We do not support libc++abi with QUANTUM_SIZE == 1'
#print >> sys.stderr, 'emcc: info: using libcxxabi, this may need CORRECT_* options'
#shared.Settings.CORRECT_SIGNS = shared.Settings.CORRECT_OVERFLOWS = shared.Settings.CORRECT_ROUNDINGS = 1
libcxxabi_symbols = map(lambda line: line.strip().split(' ')[1], open(shared.path_from_root('system', 'lib', 'libcxxabi', 'symbols')).readlines())
libcxxabi_symbols = filter(lambda symbol: symbol not in dlmalloc_symbols, libcxxabi_symbols)
libcxxabi_symbols = set(libcxxabi_symbols)
force = False # If we have libcxx, we must force inclusion of dlmalloc, since libcxx uses new internally. Note: this is kind of hacky
for name, create, fix, library_symbols in [('libcxx', create_libcxx, fix_libcxx, libcxx_symbols),
('libcxxabi', create_libcxxabi, fix_libcxxabi, libcxxabi_symbols),
('dlmalloc', create_dlmalloc, fix_dlmalloc, dlmalloc_symbols)]:
need = []
has = []
for temp_file in temp_files:
symbols = shared.Building.llvm_nm(temp_file)
for library_symbol in library_symbols:
if library_symbol in symbols.undefs:
if library_symbol in symbols.defs:
if DEBUG: print >> sys.stderr, 'emcc: considering including %s: we need |%s| and have |%s|' % (name, str(need), str(has))
if force or (need and not has):
# We need to build and link the library in
if DEBUG: print >> sys.stderr, 'emcc: including %s' % name
extra_files_to_link.append(shared.Cache.get(name, create))
force = True
if fix:
# First, combine the bitcode files if there are several. We must also link if we have a singleton .a
if len(input_files) + len(extra_files_to_link) > 1 or \
(not LEAVE_INPUTS_RAW and not (suffix(temp_files[0]) in BITCODE_SUFFIXES or suffix(temp_files[0]) in DYNAMICLIB_SUFFIXES) and shared.Building.is_ar(temp_files[0])):
linker_inputs = temp_files + extra_files_to_link
if DEBUG: print >> sys.stderr, 'emcc: linking: ', linker_inputs, in_temp(target_basename + '.bc'), remove_duplicates=remove_duplicates)
final = in_temp(target_basename + '.bc')
shutil.move(temp_files[0], in_temp(target_basename + '.bc'))
final = in_temp(target_basename + '.bc')
final = input_files[0]
print >> sys.stderr, 'emcc: saving intermediate processing steps to %s' % shared.EMSCRIPTEN_TEMP_DIR
intermediate_counter = 0
def save_intermediate(name=None, suffix='js'):
global intermediate_counter
shutil.copyfile(final, os.path.join(shared.EMSCRIPTEN_TEMP_DIR, 'emcc-%d%s.%s' % (intermediate_counter, '' if name is None else '-' + name, suffix)))
intermediate_counter += 1
if not LEAVE_INPUTS_RAW: save_intermediate('basebc', 'bc')
# Optimize, if asked to
if llvm_opts > 0 and not LEAVE_INPUTS_RAW:
if DEBUG: print >> sys.stderr, 'emcc: LLVM -O%d' % llvm_opts
shared.Building.llvm_opt(in_temp(target_basename + '.bc'), llvm_opts)
if DEBUG: save_intermediate('opt', 'bc')
# Do LTO in a separate pass to work around LLVM bug XXX (see failure e.g. in cubescript)
if llvm_lto and shared.Building.can_use_unsafe_opts() and shared.Building.can_build_standalone():
lto_opts = []
if not shared.Building.can_inline(): lto_opts.append('-disable-inlining')
if DEBUG: print >> sys.stderr, 'emcc: LLVM LTO:', lto_opts
shared.Building.llvm_opt(in_temp(target_basename + '.bc'), lto_opts)
if DEBUG: save_intermediate('lto', 'bc')
# If possible, remove dead functions etc., this potentially saves a lot in the size of the generated code (and the time to compile it)
if not LEAVE_INPUTS_RAW and shared.Building.can_build_standalone():
if DEBUG: print >> sys.stderr, 'emcc: LLVM dead globals elimination'
shared.Building.llvm_opt(in_temp(target_basename + '.bc'), ['-internalize', '-globaldce'])
if DEBUG: save_intermediate('dce', 'bc')
# Prepare .ll for Emscripten
final = shared.Building.llvm_dis(final, final + '.ll')
assert len(input_files) == 1
if DEBUG: save_intermediate('ll', 'll')
if DEBUG: print >> sys.stderr, 'emcc: autodebug'
execute(shared.ENV_PREFIX + ['python', shared.AUTODEBUGGER, final, final + '.ad.ll'])
final += '.ad.ll'
if DEBUG: save_intermediate('autodebug', 'll')
# Emscripten
if DEBUG: print >> sys.stderr, 'emcc: LLVM => JS'
extra_args = [] if not js_libraries else ['--libraries', ','.join(map(os.path.abspath, js_libraries))]
final = shared.Building.emscripten(final, append_ext=False, extra_args=extra_args)
if DEBUG: save_intermediate('original')
# Embed and preload files
if len(preload_files) + len(embed_files) > 0:
if DEBUG: print >> sys.stderr, 'emcc: setting up files'
file_args = []
if len(preload_files) > 0:
file_args += preload_files
if len(embed_files) > 0:
file_args += embed_files
if Compression.on:
file_args += ['--compress', Compression.encoder, Compression.decoder, Compression.js_name]
code = execute(shared.ENV_PREFIX + ['python', shared.FILE_PACKAGER, unsuffixed(target) + '.data'] + file_args, stdout=PIPE)[0]
src = open(final).read().replace('// {{PRE_RUN_ADDITIONS}}', code)
final += '.files.js'
open(final, 'w').write(src)
if DEBUG: save_intermediate('files')
# Apply pre and postjs files
if pre_js or post_js:
if DEBUG: print >> sys.stderr, 'emcc: applying pre/postjses'
src = open(final).read()
final += '.pp.js'
open(final, 'w').write(pre_js + src + post_js)
if DEBUG: save_intermediate('pre-post')
# Apply a source code transformation, if requested
if js_transform:
shutil.copyfile(final, final + '.tr.js')
final += '.tr.js'
posix = True if not shared.WINDOWS else False
if DEBUG: print >> sys.stderr, 'emcc: applying transform: %s' % js_transform
execute(shlex.split(js_transform, posix=posix) + [os.path.abspath(final)])
if DEBUG: save_intermediate('transformed')
# It is useful to run several js optimizer passes together, to save on unneeded unparsing/reparsing
js_optimizer_queue = []
def flush_js_optimizer_queue():
global final, js_optimizer_queue
if len(js_optimizer_queue) > 0:
if not DEBUG:
final = shared.Building.js_optimizer(final, js_optimizer_queue)
for name in js_optimizer_queue:
print >> sys.stderr, 'emcc: applying js optimization pass:', name
final = shared.Building.js_optimizer(final, [name])
js_optimizer_queue = []
if opt_level >= 1:
if DEBUG: print >> sys.stderr, 'emcc: running pre-closure post-opts'
# Clean up the syntax a bit
final = shared.Building.js_optimizer(final, [])
if DEBUG: save_intermediate('pretty')
if shared.Settings.RELOOP:
js_optimizer_queue += ['hoistMultiples', 'loopOptimizer']
if DEBUG: print >> sys.stderr, 'emcc: running variable eliminator'
final = shared.Building.eliminator(final)
if DEBUG: save_intermediate('eliminator')
js_optimizer_queue += ['simplifyExpressionsPre']
if shared.Settings.RELOOP:
js_optimizer_queue += ['optimizeShiftsAggressive'] # aggressive shifts optimization requires loops, it breaks on switches
final = shared.Building.eliminator(final) # aggressive shifts optimization introduces some new variables, remove ones that we can
if DEBUG: save_intermediate('eliminator')
if closure:
if DEBUG: print >> sys.stderr, 'emcc: running closure'
final = shared.Building.closure_compiler(final)
if DEBUG: save_intermediate('closure')
if opt_level >= 1:
if DEBUG: print >> sys.stderr, 'emcc: running post-closure post-opts'
js_optimizer_queue += ['simplifyExpressionsPost']
if minify_whitespace:
js_optimizer_queue += ['compress']
# If we were asked to also generate HTML, do that
if final_suffix == 'html':
if DEBUG: print >> sys.stderr, 'emcc: generating HTML'
shell = open(shell_path).read()
html = open(target, 'w')
if not Compression.on:
html.write(shell.replace('{{{ SCRIPT_CODE }}}', open(final).read()))
# Compress the main code
js_target = unsuffixed(target) + '.js'
shutil.move(final, js_target)
# Run the decompressor in a worker, and add code to
# 1. download the compressed file
# 2. decompress to a typed array
# 3. convert to a string of source code
# 4. insert a script element with that source code (more effective than eval)
decoding = '''
var decompressWorker = new Worker('decompress.js');
var decompressCallbacks = [];
var decompressions = 0;
Module["decompress"] = function(data, callback) {
var id = decompressCallbacks.length;
decompressWorker.postMessage({ data: data, id: id });
if (Module['setStatus']) {
decompressWorker.onmessage = function(event) {
decompressCallbacks[] = null;
if (Module['setStatus']) {
if (decompressions == 0) {
var compiledCodeXHR = new XMLHttpRequest();'GET', '%s', true);
compiledCodeXHR.responseType = 'arraybuffer';
compiledCodeXHR.onload = function() {
var arrayBuffer = compiledCodeXHR.response;
if (!arrayBuffer) throw('Loading compressed code failed.');
var byteArray = new Uint8Array(arrayBuffer);
Module.decompress(byteArray, function(decompressed) {
var source = Array.prototype.slice.apply(decompressed).map(function(x) { return String.fromCharCode(x) }).join(''); // createObjectURL instead?
var scriptTag = document.createElement('script');
scriptTag.setAttribute('type', 'text/javascript');
scriptTag.innerHTML = source;
''' % Compression.compressed_name(js_target)
html.write(shell.replace('{{{ SCRIPT_CODE }}}', decoding))
# Add decompressor with web worker glue code
decompressor = open('decompress.js', 'w')
onmessage = function(event) {
postMessage({ data: %s(, id: });
''' % Compression.js_name)
if split_js_file:
from tools.split import split_javascript_file
split_javascript_file(final, unsuffixed(target), split_js_file)
# copy final JS to output
shutil.move(final, target)
if not TEMP_DIR:
print >> sys.stderr, 'emcc saved files are in:', temp_dir
Jump to Line
Something went wrong with that request. Please try again.