#!/usr/bin/env python ''' You should normally never use this! Use emcc instead. This is a small wrapper script around the core JS compiler. This calls that compiler with the settings given to it. It can also read data from C/C++ header files (so that the JS compiler can see the constants in those headers, for the libc implementation in JS). ''' import json import optparse import os import subprocess import re import sys if not os.environ.get('EMSCRIPTEN_SUPPRESS_USAGE_WARNING'): print >> sys.stderr, ''' ============================================================== WARNING: You should normally never use this! Use emcc instead. ============================================================== ''' from tools import shared DEBUG = os.environ.get('EMCC_DEBUG') __rootpath__ = os.path.abspath(os.path.dirname(__file__)) def path_from_root(*pathelems): """Returns the absolute path for which the given path elements are relative to the emscripten root. """ return os.path.join(__rootpath__, *pathelems) temp_files = shared.TempFiles() def scan(ll, settings): # blockaddress(@main, %23) blockaddrs = [] for blockaddr in re.findall('blockaddress\([^)]*\)', ll): b = blockaddr.split('(')[1][:-1].split(', ') blockaddrs.append(b) if len(blockaddrs) > 0: settings['NECESSARY_BLOCKADDRS'] = blockaddrs def emscript(infile, settings, outfile, libraries=[]): """Runs the emscripten LLVM-to-JS compiler. We parallelize as much as possible Args: infile: The path to the input LLVM assembly file. settings: JSON-formatted settings that override the values defined in src/settings.js. outfile: The file where the output is written. """ compiler = path_from_root('src', 'compiler.js') # Parallelization: We run 3 passes: # 1 aka 'pre' : Process types and metadata and so forth, and generate the preamble. # 2 aka 'funcs': Process functions. We can parallelize this, working on each function independently. # 3 aka 'post' : Process globals, generate postamble and finishing touches. # Pre-scan ll and alter settings as necessary ll = open(infile).read() scan(ll, settings) ll = None # allow collection # Split input into the relevant parts for each phase pre = '' funcs = [] # split up functions here, for parallelism later meta = '' # needed by each function XXX post = '' in_func = False ll_lines = open(infile).readlines() for line in ll_lines: if in_func: funcs[-1] += line if line.startswith('}'): in_func = False #pre += line # XXX pre needs function defs? else: if line.startswith('define '): in_func = True funcs.append(line) #pre += line # XXX pre needs function defs? elif line.find(' = type { ') > 0: pre += line # type elif line.startswith('!'): meta += line # metadata else: post += line # global pre += line # pre needs it to, so we know about globals in pre and funcs ll_lines = None #print '========= pre ================\n' #print pre #print '========== funcs ===============\n' #for func in funcs: # print '\n// ===\n\n', func #print '========== post ==============\n' #print post #print '=========================\n' # Save settings to a file to work around v8 issue 1579 settings_file = temp_files.get('.txt').name s = open(settings_file, 'w') s.write(json.dumps(settings)) s.close() # Pass 1 if DEBUG: print >> sys.stderr, 'phase 1' pre_file = temp_files.get('.ll').name open(pre_file, 'w').write(pre) out = shared.run_js(compiler, shared.COMPILER_ENGINE, [settings_file, pre_file, 'pre'] + libraries, stdout=subprocess.PIPE, cwd=path_from_root('src')) js, forwarded_data = out.split('//FORWARDED_DATA:') #print 'js', js #print >> sys.stderr, 'FORWARDED_DATA 1:', forwarded_data, type(forwarded_data) forwarded_file = temp_files.get('.json').name open(forwarded_file, 'w').write(forwarded_data) # Pass 2 if DEBUG: print >> sys.stderr, 'phase 2' funcs_file = temp_files.get('.ll').name open(funcs_file, 'w').write('\n'.join(funcs) + '\n' + meta) #print 'pass 2c..'#, open(funcs_file).read() out = shared.run_js(compiler, shared.COMPILER_ENGINE, [settings_file, funcs_file, 'funcs', forwarded_file] + libraries, stdout=subprocess.PIPE, cwd=path_from_root('src')) funcs_js, forwarded_data = out.split('//FORWARDED_DATA:') #print 'js', js forwarded_file += '2' #print >> sys.stderr, 'FORWARDED_DATA 2:', forwarded_data, type(forwarded_data), forwarded_file open(forwarded_file, 'w').write(forwarded_data) # XXX must coordinate function indexixing data when parallelizing #print 'OUT\n', out js += funcs_js # Pass 3 if DEBUG: print >> sys.stderr, 'phase 3' post_file = temp_files.get('.ll').name open(post_file, 'w').write(post) out = shared.run_js(compiler, shared.COMPILER_ENGINE, [settings_file, post_file, 'post', forwarded_file] + libraries, stdout=subprocess.PIPE, cwd=path_from_root('src')) js += out outfile.write(js) outfile.close() def main(args): # Prepare settings for serialization to JSON. settings = {} for setting in args.settings: name, value = setting.strip().split('=', 1) settings[name] = json.loads(value) # Add header defines to settings defines = {} include_root = path_from_root('system', 'include') headers = args.headers[0].split(',') if len(args.headers) > 0 else [] seen_headers = set() while len(headers) > 0: header = headers.pop(0) if not os.path.isabs(header): header = os.path.join(include_root, header) seen_headers.add(header) for line in open(header, 'r'): line = line.replace('\t', ' ') m = re.match('^ *# *define +(?P[-\w_.]+) +\(?(?P[-\w_.|]+)\)?.*', line) if not m: # Catch enum defines of a very limited sort m = re.match('^ +(?P[A-Z_\d]+) += +(?P\d+).*', line) if m: if m.group('name') != m.group('value'): defines[m.group('name')] = m.group('value') #else: # print 'Warning: %s #defined to itself' % m.group('name') # XXX this can happen if we are set to be equal to an enum (with the same name) m = re.match('^ *# *include *["<](?P[\w_.-/]+)[">].*', line) if m: # Find this file found = False for w in [w for w in os.walk(include_root)]: for f in w[2]: curr = os.path.join(w[0], f) if curr.endswith(m.group('name')) and curr not in seen_headers: headers.append(curr) found = True break if found: break #assert found, 'Could not find header: ' + m.group('name') if len(defines) > 0: def lookup(value): try: while not unicode(value).isnumeric(): value = defines[value] return value except: pass try: # 0x300 etc. value = eval(value) return value except: pass try: # CONST1|CONST2 parts = map(lookup, value.split('|')) value = reduce(lambda a, b: a|b, map(eval, parts)) return value except: pass return None for key, value in defines.items(): value = lookup(value) if value is not None: defines[key] = str(value) else: del defines[key] #print >> sys.stderr, 'new defs:', str(defines).replace(',', ',\n '), '\n\n' settings.setdefault('C_DEFINES', {}).update(defines) # libraries libraries = args.libraries[0].split(',') if len(args.libraries) > 0 else [] # Compile the assembly to Javascript. emscript(args.infile, settings, args.outfile, libraries) if __name__ == '__main__': parser = optparse.OptionParser( usage='usage: %prog [-h] [-H HEADERS] [-o OUTFILE] [-s FOO=BAR]* infile', description=('You should normally never use this! Use emcc instead. ' 'This is a wrapper around the JS compiler, converting .ll to .js.'), epilog='') parser.add_option('-H', '--headers', default=[], action='append', help='System headers (comma separated) whose #defines should be exposed to the compiled code.') parser.add_option('-L', '--libraries', default=[], action='append', help='Library files (comma separated) to use in addition to those in emscripten src/library_*.') parser.add_option('-o', '--outfile', default=sys.stdout, help='Where to write the output; defaults to stdout.') parser.add_option('-s', '--setting', dest='settings', default=[], action='append', metavar='FOO=BAR', help=('Overrides for settings defined in settings.js. ' 'May occur multiple times.')) # Convert to the same format that argparse would have produced. keywords, positional = parser.parse_args() if len(positional) != 1: raise RuntimeError('Must provide exactly one positional argument.') keywords.infile = os.path.abspath(positional[0]) if isinstance(keywords.outfile, basestring): keywords.outfile = open(keywords.outfile, 'w') temp_files.run_and_clean(lambda: main(keywords))