diff options
author | Julian Smith <jules@op59.net> | 2020-05-15 11:41:30 +0100 |
---|---|---|
committer | Julian Smith <jules@op59.net> | 2020-06-18 13:03:00 +0100 |
commit | e29a4c0affe3e7b66ff19d70e0d2abbc1e7fdf21 (patch) | |
tree | 5e63d4e4a4cb4acd3f83af12c7601fd50076997d /toolbin | |
parent | 31157290172feea1817d6e69c5064d025a9ab45c (diff) | |
download | ghostpdl-e29a4c0affe3e7b66ff19d70e0d2abbc1e7fdf21.tar.gz |
Added experimental swig wrapper for gsapi.
Diffstat (limited to 'toolbin')
-rwxr-xr-x | toolbin/gsapiwrap.py | 699 | ||||
-rw-r--r-- | toolbin/jlib.py | 1354 |
2 files changed, 2053 insertions, 0 deletions
diff --git a/toolbin/gsapiwrap.py b/toolbin/gsapiwrap.py new file mode 100755 index 000000000..0ef0bb088 --- /dev/null +++ b/toolbin/gsapiwrap.py @@ -0,0 +1,699 @@ +#! /usr/bin/env python3 + +''' +Use Swig to build wrappers for gsapi. + +Example usage: + + Note that we use mupdf's scripts/jlib.py, and assume that there is a mupdf + checkout in the parent directory of the ghostpdl checkout - see 'import + jlib' below. + + ./toolbin/gsapiwrap.py --python -l -0 -1 -t + Build python wrapper for gsapi and run simple test. + + ./toolbin/gsapiwrap.py --csharp -l -0 -1 -t + Build C# wrapper for gsapi and run simple test. + +Args: + + -c: + Clean language-specific out-dir. + + -l: + Build libgs.so (by running make). + + -0: + Run swig to generate language-specific files. + + -1: + Generate language wrappers by compiling/linking the files generated by + -0. + + --csharp: + Generate C# wrappers (requires Mono on Linux). Should usually be first + param. + + --python + Generate Python wrappers. Should usually be first param. + + --swig <swig> + Set location of swig binary. + + -t + Run simple test of language wrappers generated by -1. + +Status: + As of 2020-05-22: + Some python wrappers seem to work ok. + + C# wrappers are not implemented for gsapi_set_poll() and + gsapi_set_stdio(). +''' + +import os +import re +import sys +import textwrap + +import jlib + + +def devpython_info(): + ''' + Use python3-config to find libpython.so and python-dev include path etc. + ''' + python_configdir = jlib.system( 'python3-config --configdir', out='return') + libpython_so = os.path.join( + python_configdir.strip(), + f'libpython{sys.version_info[0]}.{sys.version_info[1]}.so', + ) + assert os.path.isfile( libpython_so), f'cannot find libpython_so={libpython_so}' + + python_includes = jlib.system( 'python3-config --includes', out='return') + python_includes = python_includes.strip() + return python_includes, libpython_so + +def swig_version( swig='swig'): + t = jlib.system( f'{swig} -version', out='return') + m = re.search( 'SWIG Version ([0-9]+)[.]([0-9]+)[.]([0-9]+)', t) + assert m + swig_major = int( m.group(1)) + return swig_major + + +dir_ghostpdl = os.path.abspath( f'{__file__}/../../') + '/' + + +def out_dir( language): + if language == 'python': + return 'gsapiwrap/python/' + if language == 'csharp': + return 'gsapiwrap/csharp/' + assert 0 + +def out_so( language): + ''' + Returns name of .so that implements language-specific wrapper. I think + these names have to match what the language runtime requires. + + For python, Swig generates a module foo.py which does 'import _foo'. + + Similarly C# assumes a file called 'libfoo.so'. + ''' + if language == 'python': + return f'{out_dir(language)}_gsapi.so' + if language == 'csharp': + return f'{out_dir(language)}libgsapi.so' + assert 0 + +def lib_gs_info(): + return f'{dir_ghostpdl}sodebugbin/libgs.so', 'make sodebug' + return f'{dir_ghostpdl}sobin/libgs.so', 'make so' + +def lib_gs(): + ''' + Returns name of the gs shared-library. + ''' + return lib_gs_info()[0] + + +def swig_i( swig, language): + ''' + Returns text for a swig .i file for psi/iapi.h. + ''' + swig_major = swig_version( swig) + + + # We need to redeclare or wrap some functions, e.g. to add OUTPUT + # annotations. We use #define, %ignore and #undef to hide the original + # declarations in the .h file. + # + fns_redeclare = ( + 'gsapi_run_file', + 'gsapi_run_string', + 'gsapi_run_string_begin', + 'gsapi_run_string_continue', + 'gsapi_run_string_end', + 'gsapi_run_string_with_length', + 'gsapi_set_poll', + 'gsapi_set_poll_with_handle', + 'gsapi_set_stdio', + 'gsapi_set_stdio_with_handle', + 'gsapi_new_instance', + ) + + + swig_i_text = textwrap.dedent(f''' + %module(directors="1") gsapi + + %include cpointer.i + %pointer_functions(int, pint); + + // This seems to be necessary to make csharp handle OUTPUT args. + // + %include typemaps.i + + // For gsapi_init_with_args(). + %include argcargv.i + + %include cstring.i + + // Include type information in python doc strings. If we have + // swig-4, we can propogate comments from the C api instead, which + // is preferred. + // + {'%feature("autodoc", "3");' if swig_major < 4 else ''} + + %{{ + #include "psi/iapi.h" + //#include "base/gserrors.h" + + // Define wrapper functions that present a modified API that + // swig can cope with. + // + + // Swig cannot handle void** out-param. + // + static void* new_instance( void* caller_handle, int* out) + {{ + void* ret = NULL; + *out = gsapi_new_instance( &ret, caller_handle); + printf( "gsapi_new_instance() returned *out=%i ret=%p\\n", *out, ret); + fflush( stdout); + return ret; + }} + + // Swig cannot handle (const char* str, int strlen) args. + // + static int run_string_continue(void *instance, const char *str, int user_errors, int *pexit_code) {{ + + return gsapi_run_string_continue( instance, str, strlen(str), user_errors, pexit_code); + }} + %}} + + // Strip gsapi_ prefix from all generated names. + // + %rename("%(strip:[gsapi_])s") ""; + + // Tell Swig about gsapi_get_default_device_list()'s out-params, so + // it adds them to the returned object. + // + // I think the '(void) *$1' will ensure that swig code doesn't + // attempt to free() the returned string. + // + {'%cstring_output_allocate_size(char **list, int *listlen, (void) *$1);' if language == 'python' else ''} + + // Tell swig about the (argc,argv) args in gsapi_init_with_args(). + // + %apply (int ARGC, char **ARGV) {{ (int argc, char **argv) }} + + // Support for wrapping various functions that take function + // pointer args. For each, we define a wrapper function that, + // instead of having function pointer args, takes a class with + // virtual methods. This allows swig to wrap things - python/c# etc + // can create a derived class that implements these virtual methods + // in the python/c# world. + // + + // Wrap gsapi_set_stdio_with_handle(). + // + %feature("director") set_stdio_class; + + %inline {{ + struct set_stdio_class {{ + + virtual int stdin_fn( char* buf, int len) = 0; + virtual int stdout_fn( const char* buf, int len) = 0; + virtual int stderr_fn( const char* buf, int len) = 0; + + static int stdin_fn_wrap( void *caller_handle, char *buf, int len) {{ + return ((set_stdio_class*) caller_handle)->stdin_fn(buf, len); + }} + static int stdout_fn_wrap( void *caller_handle, const char *buf, int len) {{ + return ((set_stdio_class*) caller_handle)->stdout_fn(buf, len); + }} + static int stderr_fn_wrap( void *caller_handle, const char *buf, int len) {{ + return ((set_stdio_class*) caller_handle)->stderr_fn(buf, len); + }} + + virtual ~set_stdio_class() {{}} + }}; + + int set_stdio_with_class( void *instance, set_stdio_class* class_) {{ + return gsapi_set_stdio_with_handle( + instance, + set_stdio_class::stdin_fn_wrap, + set_stdio_class::stdout_fn_wrap, + set_stdio_class::stderr_fn_wrap, + (void*) class_ + ); + }} + + + }} + + // Wrap gsapi_set_poll(). + // + %feature("director") set_poll_class; + + %inline {{ + struct set_poll_class {{ + virtual int poll_fn() = 0; + + static int poll_fn_wrap( void* caller_handle) {{ + return ((set_poll_class*) caller_handle)->poll_fn(); + }} + + virtual ~set_poll_class() {{}} + }}; + + int set_poll_with_class( void* instance, set_poll_class* class_) {{ + return gsapi_set_poll_with_handle( + instance, + set_poll_class::poll_fn_wrap, + (void*) class_ + ); + }} + + }} + + // For functions that we re-declare (typically to specify OUTPUT on + // one or more args), use a macro to rename the declaration in the + // header file and tell swig to ignore these renamed declarations. + // + ''') + + for fn in fns_redeclare: + swig_i_text += f'#define {fn} {fn}0\n' + + for fn in fns_redeclare: + swig_i_text += f'%ignore {fn}0;\n' + + swig_i_text += textwrap.dedent(f''' + #include "psi/iapi.h" + //#include "base/gserrors.h" + ''') + + for fn in fns_redeclare: + swig_i_text += f'#undef {fn}\n' + + + swig_i_text += textwrap.dedent(f''' + // Tell swig about our wrappers and altered declarations. + // + + // Use swig's OUTPUT annotation for out-parameters. + // + int gsapi_run_file(void *instance, const char *file_name, int user_errors, int *OUTPUT); + int gsapi_run_string_begin(void *instance, int user_errors, int *OUTPUT); + int gsapi_run_string_end(void *instance, int user_errors, int *OUTPUT); + //int gsapi_run_string_with_length(void *instance, const char *str, unsigned int length, int user_errors, int *OUTPUT); + int gsapi_run_string(void *instance, const char *str, int user_errors, int *OUTPUT); + + // Declare functions defined above that we want swig to wrap. These + // don't have the gsapi_ prefix, so that they can internally call + // the wrapped gsapi_*() function. [We've told swig to strip the + // gsapi_ prefix on generated functions anyway, so this doesn't + // afffect the generated names.] + // + static int run_string_continue(void *instance, const char *str, int user_errors, int *OUTPUT); + static void* new_instance(void* caller_handle, int* OUTPUT); + ''') + + if language == 'python': + swig_i_text += textwrap.dedent(f''' + + // Define python code that is needed to handle functions with + // function-pointer args. + // + %pythoncode %{{ + + set_stdio_g = None + def set_stdio( instance, stdin, stdout, stderr): + class derived( set_stdio_class): + def stdin_fn( self): + return stdin() + def stdout_fn( self, text, len): + return stdout( text, len) + def stderr_fn( self, text, len): + return stderr( text) + + global set_stdio_g + set_stdio_g = derived() + return set_stdio_with_class( instance, set_stdio_g) + + set_poll_g = None + def set_poll( instance, fn): + class derived( set_poll_class): + def poll_fn( self): + return fn() + global set_poll_g + set_poll_g = derived() + return set_poll_with_class( instance, set_poll_g) + %}} + ''') + + return swig_i_text + + + +def run_swig( swig, language): + ''' + Runs swig using a generated .i file. + + The .i file modifies the gsapi API in places to allow specification of + out-parameters that swig understands - e.g. void** OUTPUT doesn't work. + ''' + os.makedirs( out_dir(language), exist_ok=True) + swig_major = swig_version( swig) + + swig_i_text = swig_i( swig, language) + swig_i_filename = f'{out_dir(language)}iapi.i' + jlib.update_file( swig_i_text, swig_i_filename) + + out_cpp = f'{out_dir(language)}gsapi.cpp' + + if language == 'python': + out_lang = f'{out_dir(language)}gsapi.py' + elif language == 'csharp': + out_lang = f'{out_dir(language)}gsapi.cs' + else: + assert 0 + + out_files = (out_cpp, out_lang) + + doxygen_arg = '' + if swig_major >= 4 and language == 'python': + doxygen_arg = '-doxygen' + + extra = '' + if language == 'csharp': + # Tell swig to put all generated csharp code into a single file. + extra = f'-outfile gsapi.cs' + + command = (textwrap.dedent(f''' + {swig} + -Wall + -c++ + -{language} + {doxygen_arg} + -module gsapi + -outdir {out_dir(language)} + -o {out_cpp} + {extra} + -includeall + -I{dir_ghostpdl} + -ignoremissing + {swig_i_filename} + ''').strip().replace( '\n', ' \\\n') + ) + + jlib.build( + (swig_i_filename,), + out_files, + command, + prefix=' ', + ) + + +def main( argv): + + swig = 'swig' + language = 'python' + + args = jlib.Args( sys.argv[1:]) + while 1: + try: + arg = args.next() + except StopIteration: + break + + if 0: + pass + + elif arg == '-c': + jlib.system( f'rm {out_dir(language)}* || true', verbose=1, prefix=' ') + + elif arg == '-l': + command = lib_gs_info()[1] + jlib.system( command, verbose=1, prefix=' ') + + elif arg == '-0': + run_swig( swig, language) + + elif arg == '-1': + + libs = [lib_gs()] + includes = [dir_ghostpdl] + file_cpp = f'{out_dir(language)}gsapi.cpp' + + if language == 'python': + python_includes, libpython_so = devpython_info() + libs.append( libpython_so) + includes.append( python_includes) + + includes_text = '' + for i in includes: + includes_text += f' -I{i}' + command = textwrap.dedent(f''' + g++ + -g + -Wall -W + -o {out_so(language)} + -fPIC + -shared + {includes_text} + {jlib.link_l_flags(libs)} + {file_cpp} + ''').strip().replace( '\n', ' \\\n') + jlib.build( + (file_cpp, lib_gs(), 'psi/iapi.h'), + (out_so(language),), + command, + prefix=' ', + ) + + elif arg == '--csharp': + language = 'csharp' + + elif arg == '--python': + language = 'python' + + elif arg == '--swig': + swig = args.next() + + elif arg == '-t': + + if language == 'python': + text = textwrap.dedent(''' + #!/usr/bin/env python3 + + import os + import sys + + import gsapi + + gsapi.gs_error_Quit = -101 + + def main(): + minst, code = gsapi.new_instance(None) + print( f'minst={minst} code={code}') + + if 1: + def stdin_local(len): + # Not sure whether this is right. + return sys.stdin.read(len) + def stdout_local(text, l): + sys.stdout.write(text[:l]) + return l + def stderr_local(text, l): + sys.stderr.write(text[:l]) + return l + gsapi.set_stdio( minst, None, stdout_local, stderr_local); + + if 1: + def poll_fn(): + return 0 + gsapi.set_poll(minst, poll_fn) + if 1: + s = 'display x11alpha x11 bbox' + gsapi.set_default_device_list( minst, s, len(s)) + + e, text = gsapi.get_default_device_list( minst) + print( f'gsapi.get_default_device_list() returned e={e} text={text!r}') + + out = 'out.pdf' + if os.path.exists( out): + os.remove( out) + assert not os.path.exists( out) + + gsargv = [''] + gsargv += f'-dNOPAUSE -dBATCH -dSAFER -sDEVICE=pdfwrite -sOutputFile={out} contrib/pcl3/ps/levels-test.ps'.split() + print( f'gsargv={gsargv}') + code = gsapi.set_arg_encoding(minst, gsapi.GS_ARG_ENCODING_UTF8) + if code == 0: + code = gsapi.init_with_args(minst, gsargv) + + code, exit_code = gsapi.run_string_begin( minst, 0) + print( f'gsapi.run_string_begin() returned code={code} exit_code={exit_code}') + assert code == 0 + assert exit_code == 0 + + gsapi.run_string + + code1 = gsapi.exit(minst) + if (code == 0 or code == gsapi.gs_error_Quit): + code = code1 + gsapi.delete_instance(minst) + assert os.path.isfile( out) + if code == 0 or code == gsapi.gs_error_Quit: + return 0 + return 1 + + if __name__ == '__main__': + code = main() + assert code == 0 + sys.exit( code) + ''') + text = text[1:] # skip leading \n. + test_py = f'{out_dir(language)}test.py' + jlib.update_file( text, test_py) + os.chmod( test_py, 0o744) + + jlib.system( + f'LD_LIBRARY_PATH={os.path.abspath( f"{lib_gs()}/..")}' + f' PYTHONPATH={out_dir(language)}' + f' {test_py}' + , + verbose = 1, + prefix=' ', + ) + + elif language == 'csharp': + # See: https://github.com/swig/swig/blob/master/Lib/csharp/typemaps.i + # + text = textwrap.dedent(''' + using System; + public class runme { + static void Main() { + int code; + SWIGTYPE_p_void instance; + Console.WriteLine("hello world"); + instance = gsapi.new_instance(null, out code); + Console.WriteLine("code is: " + code); + gsapi.add_control_path(instance, 0, "hello"); + } + } + ''') + test_cs = f'{out_dir(language)}test.cs' + jlib.update_file( text, test_cs) + files_in = f'{out_dir(language)}gsapi.cs', test_cs + file_out = f'{out_dir(language)}test.exe' + command = f'mono-csc -debug+ -out:{file_out} {" ".join(files_in)}' + jlib.build( files_in, (file_out,), command, prefix=' ') + + ld_library_path = f'{dir_ghostpdl}sobin' + jlib.system( f'LD_LIBRARY_PATH={ld_library_path} {file_out}', verbose=jlib.log, prefix=' ') + + elif arg == '--tt': + # small swig test case. + os.makedirs( 'swig-tt', exist_ok=True) + i = textwrap.dedent(f''' + %include cpointer.i + %include cstring.i + %feature("autodoc", "3"); + %cstring_output_allocate_size(char **list, int *listlen, (void) *$1); + %inline {{ + static inline int gsapi_get_default_device_list(void *instance, char **list, int *listlen) + {{ + *list = (char*) "hello world"; + *listlen = 6; + return 0; + }} + }} + ''') + jlib.update_file(i, 'swig-tt/tt.i') + jlib.system('swig -c++ -python -module tt -outdir swig-tt -o swig-tt/tt.cpp swig-tt/tt.i', verbose=1) + p = textwrap.dedent(f''' + #!/usr/bin/env python3 + import tt + print( tt.gsapi_get_default_device_list(None)) + ''')[1:] + jlib.update_file( p, 'swig-tt/test.py') + python_includes, python_so = devpython_info() + includes = f'-I {python_includes}' + link_flags = jlib.link_l_flags( [python_so]) + jlib.system( f'g++ -shared -fPIC {includes} {link_flags} -o swig-tt/_tt.so swig-tt/tt.cpp', verbose=1) + jlib.system( f'cd swig-tt; python3 test.py', verbose=1) + + elif arg == '-T': + # Very simple test that we can create c# wrapper for trivial code. + os.makedirs( 'swig-cs-test', exist_ok=True) + example_cpp = textwrap.dedent(''' + #include <time.h> + double My_variable = 3.0; + + int fact(int n) { + if (n <= 1) return 1; + else return n*fact(n-1); + } + + int my_mod(int x, int y) { + return (x%y); + } + + char *get_time() + { + time_t ltime; + time(<ime); + return ctime(<ime); + } + ''') + jlib.update_file( example_cpp, 'swig-cs-test/example.cpp') + + example_i = textwrap.dedent(''' + %module example + %{ + /* Put header files here or function declarations like below */ + extern double My_variable; + extern int fact(int n); + extern int my_mod(int x, int y); + extern char *get_time(); + %} + + extern double My_variable; + extern int fact(int n); + extern int my_mod(int x, int y); + extern char *get_time(); + ''') + jlib.update_file( example_i, 'swig-cs-test/example.i') + + runme_cs = textwrap.dedent(''' + using System; + public class runme { + static void Main() { + Console.WriteLine(example.My_variable); + Console.WriteLine(example.fact(5)); + Console.WriteLine(example.get_time()); + } + } + ''') + jlib.update_file( runme_cs, 'swig-cs-test/runme.cs') + jlib.system( 'g++ -g -fPIC -shared -o swig-cs-test/libfoo.so swig-cs-test/example.cpp', verbose=1) + jlib.system( 'swig -c++ -csharp -module example -outdir swig-cs-test -o swig-cs-test/example_wrap.cpp -outfile example.cs swig-cs-test/example.i', verbose=1) + jlib.system( 'g++ -g -fPIC -shared -L swig-cs-test -l foo swig-cs-test/example_wrap.cpp -o swig-cs-test/libexample.so', verbose=1) + jlib.system( 'cd swig-cs-test; mono-csc -out:runme.exe example.cs runme.cs', verbose=1) + jlib.system( 'cd swig-cs-test; LD_LIBRARY_PATH=`pwd` ./runme.exe', verbose=1) + jlib.system( 'ls -l swig-cs-test', verbose=1) + + + else: + raise Exception( f'unrecognised arg: {arg}') + +if __name__ == '__main__': + try: + main( sys.argv) + except Exception as e: + jlib.exception_info( out=sys.stdout) + sys.exit(1) diff --git a/toolbin/jlib.py b/toolbin/jlib.py new file mode 100644 index 000000000..54ee0a222 --- /dev/null +++ b/toolbin/jlib.py @@ -0,0 +1,1354 @@ +from __future__ import print_function + +import codecs +import inspect +import io +import os +import shutil +import subprocess +import sys +import time +import traceback + + +def place( frame_record): + ''' + Useful debugging function - returns representation of source position of + caller. + ''' + filename = frame_record.filename + line = frame_record.lineno + function = frame_record.function + ret = os.path.split( filename)[1] + ':' + str( line) + ':' + function + ':' + if 0: + tid = str( threading.currentThread()) + ret = '[' + tid + '] ' + ret + return ret + + +def expand_nv( text, caller): + ''' + Returns <text> with special handling of {<expression>} items. + + text: + String containing {<expression>} items. + caller: + If an int, the number of frames to step up when looking for file:line + information or evaluating expressions. + + Otherwise should be a frame record as returned by inspect.stack()[]. + + <expression> is evaluated in <caller>'s context using eval(), and expanded + to <expression> or <expression>=<value>. + + If <expression> ends with '=', this character is removed and we prefix the + result with <expression>=. + + E.g.: + x = 45 + y = 'hello' + expand_nv( 'foo {x} {y=}') + returns: + foo 45 y=hello + + <expression> can also use ':' and '!' to control formatting, like + str.format(). + ''' + if isinstance( caller, int): + frame_record = inspect.stack()[ caller] + else: + frame_record = caller + frame = frame_record.frame + try: + def get_items(): + ''' + Yields (pre, item), where <item> is contents of next {...} or None, + and <pre> is preceding text. + ''' + pos = 0 + pre = '' + while 1: + if pos == len( text): + yield pre, None + break + rest = text[ pos:] + if rest.startswith( '{{') or rest.startswith( '}}'): + pre += rest[0] + pos += 2 + elif text[ pos] == '{': + close = text.find( '}', pos) + if close < 0: + raise Exception( 'After "{" at offset %s, cannot find closing "}". text is: %r' % ( + pos, text)) + yield pre, text[ pos+1 : close] + pre = '' + pos = close + 1 + else: + pre += text[ pos] + pos += 1 + + ret = '' + for pre, item in get_items(): + ret += pre + nv = False + if item: + if item.endswith( '='): + nv = True + item = item[:-1] + expression, tail = split_first_of( item, '!:') + try: + value = eval( expression, frame.f_globals, frame.f_locals) + value_text = ('{0%s}' % tail).format( value) + except Exception as e: + value_text = '{??Failed to evaluate %r in context %s:%s because: %s??}' % ( + expression, + frame_record.filename, + frame_record.lineno, + e, + ) + if nv: + ret += '%s=' % expression + ret += value_text + + return ret + + finally: + del frame + + +class LogPrefixTime: + def __init__( self, date=False, time_=True, elapsed=False): + self.date = date + self.time = time_ + self.elapsed = elapsed + self.t0 = time.time() + def __call__( self): + ret = '' + if self.date: + ret += time.strftime( ' %F') + if self.time: + ret += time.strftime( ' %T') + if self.elapsed: + ret += ' (+%s)' % time_duration( time.time() - self.t0, s_format='%.1f') + if ret: + ret = ret.strip() + ': ' + return ret + +class LogPrefixFileLine: + def __call__( self, caller): + if isinstance( caller, int): + caller = inspect.stack()[ caller] + return place( caller) + ' ' + +class LogPrefixScopes: + ''' + Internal use only. + ''' + def __init__( self): + self.items = [] + def __call__( self): + ret = '' + for item in self.items: + if callable( item): + item = item() + ret += item + return ret + + +class LogPrefixScope: + ''' + Can be used to insert scoped prefix to log output. + ''' + def __init__( self, prefix): + g_log_prefixe_scopes.items.append( prefix) + def __enter__( self): + pass + def __exit__( self, exc_type, exc_value, traceback): + global g_log_prefix + g_log_prefixe_scopes.items.pop() + + +g_log_delta = 0 + +class LogDeltaScope: + ''' + Can be used to temporarily change verbose level of logging. + + E.g to temporarily increase logging: + + with jlib.LogDeltaScope(-1): + ... + ''' + def __init__( self, delta): + self.delta = delta + global g_log_delta + g_log_delta += self.delta + def __enter__( self): + pass + def __exit__( self, exc_type, exc_value, traceback): + global g_log_delta + g_log_delta -= self.delta + +# Special item that can be inserted into <g_log_prefixes> to enable +# temporary addition of text into log prefixes. +# +g_log_prefixe_scopes = LogPrefixScopes() + +# List of items that form prefix for all output from log(). +# +g_log_prefixes = [] + + +def log_text( text=None, caller=1, nv=True): + ''' + Returns log text, prepending all lines with text from g_log_prefixes. + + text: + The text to output. Each line is prepended with prefix text. + caller: + If an int, the number of frames to step up when looking for file:line + information or evaluating expressions. + + Otherwise should be a frame record as returned by inspect.stack()[]. + nv: + If true, we expand {...} in <text> using expand_nv(). + ''' + if isinstance( caller, int): + caller += 1 + prefix = '' + for p in g_log_prefixes: + if callable( p): + if isinstance( p, LogPrefixFileLine): + p = p(caller) + else: + p = p() + prefix += p + + if text is None: + return prefix + + if nv: + text = expand_nv( text, caller) + + if text.endswith( '\n'): + text = text[:-1] + lines = text.split( '\n') + + text = '' + for line in lines: + text += prefix + line + '\n' + return text + + + +s_log_levels_cache = dict() +s_log_levels_items = [] + +def log_levels_find( caller): + if not s_log_levels_items: + return 0 + + tb = traceback.extract_stack( None, 1+caller) + if len(tb) == 0: + return 0 + filename, line, function, text = tb[0] + + key = function, filename, line, + delta = s_log_levels_cache.get( key) + + if delta is None: + # Calculate and populate cache. + delta = 0 + for item_function, item_filename, item_delta in s_log_levels_items: + if item_function and not function.startswith( item_function): + continue + if item_filename and not filename.startswith( item_filename): + continue + delta = item_delta + break + + s_log_levels_cache[ key] = delta + + return delta + + +def log_levels_add( delta, filename_prefix, function_prefix): + ''' + log() calls from locations with filenames starting with <filename_prefix> + and/or function names starting with <function_prefix> will have <delta> + added to their level. + + Use -ve delta to increase verbosity from particular filename or function + prefixes. + ''' + log( 'adding level: {filename_prefix=!r} {function_prefix=!r}') + + # Sort in reverse order so that long functions and filename specs come + # first. + # + s_log_levels_items.append( (function_prefix, filename_prefix, delta)) + s_log_levels_items.sort( reverse=True) + + +def log( text, level=0, caller=1, nv=True, out=None): + ''' + Writes log text, with special handling of {<expression>} items in <text> + similar to python3's f-strings. + + text: + The text to output. + caller: + How many frames to step up to get caller's context when evaluating + file:line information and/or expressions. Or frame record as returned + by inspect.stack()[]. + nv: + If true, we expand {...} in <text> using expand_nv(). + out: + Where to send output. If None we use sys.stdout. + + <expression> is evaluated in our caller's context (<n> stack frames up) + using eval(), and expanded to <expression> or <expression>=<value>. + + If <expression> ends with '=', this character is removed and we prefix the + result with <expression>=. + + E.g.: + x = 45 + y = 'hello' + expand_nv( 'foo {x} {y=}') + returns: + foo 45 y=hello + + <expression> can also use ':' and '!' to control formatting, like + str.format(). + ''' + if out is None: + out = sys.stdout + level += g_log_delta + if isinstance( caller, int): + caller += 1 + level += log_levels_find( caller) + if level <= 0: + text = log_text( text, caller, nv=nv) + out.write( text) + out.flush() + + +def log0( text, caller=1, nv=True, out=None): + ''' + Most verbose log. Same as log(). + ''' + log( text, level=0, caller=caller+1, nv=nv, out=out) + +def log1( text, caller=1, nv=True, out=None): + log( text, level=1, caller=caller+1, nv=nv, out=out) + +def log2( text, caller=1, nv=True, out=None): + log( text, level=2, caller=caller+1, nv=nv, out=out) + +def log3( text, caller=1, nv=True, out=None): + log( text, level=3, caller=caller+1, nv=nv, out=out) + +def log4( text, caller=1, nv=True, out=None): + log( text, level=4, caller=caller+1, nv=nv, out=out) + +def log5( text, caller=1, nv=True, out=None): + ''' + Least verbose log. + ''' + log( text, level=5, caller=caller+1, nv=nv, out=out) + +def logx( text, caller=1, nv=True, out=None): + ''' + Does nothing, useful when commenting out a log(). + ''' + pass + + +def log_levels_add_env( name='JLIB_log_levels'): + ''' + Added log levels encoded in an environmental variable. + ''' + t = os.environ.get( name) + if t: + for ffll in t.split( ','): + ffl, delta = ffll.split( '=', 1) + delta = int( delta) + ffl = ffl.split( ':') + if 0: + pass + elif len( ffl) == 1: + filename = ffl + function = None + elif len( ffl) == 2: + filename, function = ffl + else: + assert 0 + log_levels_add( delta, filename, function) + + +def strpbrk( text, substrings): + ''' + Finds first occurrence of any item in <substrings> in <text>. + + Returns (pos, substring) or (len(text), None) if not found. + ''' + ret_pos = len( text) + ret_substring = None + for substring in substrings: + pos = text.find( substring) + if pos >= 0 and pos < ret_pos: + ret_pos = pos + ret_substring = substring + return ret_pos, ret_substring + + +def split_first_of( text, substrings): + ''' + Returns (pre, post), where <pre> doesn't contain any item in <substrings> + and <post> is empty or starts with an item in <substrings>. + ''' + pos, _ = strpbrk( text, substrings) + return text[ :pos], text[ pos:] + + + +log_levels_add_env() + + +def force_line_buffering(): + ''' + Ensure sys.stdout and sys.stderr are line-buffered. E.g. makes things work + better if output is piped to a file via 'tee'. + + Returns original out,err streams. + ''' + stdout0 = sys.stdout + stderr0 = sys.stderr + sys.stdout = os.fdopen( os.dup( sys.stdout.fileno()), 'w', 1) + sys.stderr = os.fdopen( os.dup( sys.stderr.fileno()), 'w', 1) + return stdout0, stderr0 + + +def exception_info( exception=None, limit=None, out=None, prefix='', oneline=False): + ''' + General replacement for traceback.* functions that print/return information + about exceptions. This function provides a simple way of getting the + functionality provided by these traceback functions: + + traceback.format_exc() + traceback.format_exception() + traceback.print_exc() + traceback.print_exception() + + Returns: + A string containing description of specified exception and backtrace. + + Inclusion of outer frames: + We improve upon traceback.* in that we also include stack frames above + the point at which an exception was caught - frames from the top-level + <module> or thread creation fn to the try..catch block, which makes + backtraces much more useful. + + Google 'sys.exc_info backtrace incomplete' for more details. + + We deliberately leave a slightly curious pair of items in the backtrace + - the point in the try: block that ended up raising an exception, and + the point in the associated except: block from which we were called. + + For clarity, we insert an empty frame in-between these two items, so + that one can easily distinguish the two parts of the backtrace. + + So the backtrace looks like this: + + root (e.g. <module> or /usr/lib/python2.7/threading.py:778:__bootstrap(): + ... + file:line in the except: block where the exception was caught. + ::(): marker + file:line in the try: block. + ... + file:line where the exception was raised. + + The items after the ::(): marker are the usual items that traceback.* + shows for an exception. + + Also the backtraces that are generated are more concise than those provided + by traceback.* - just one line per frame instead of two - and filenames are + output relative to the current directory if applicatble. And one can easily + prefix all lines with a specified string, e.g. to indent the text. + + Returns a string containing backtrace and exception information, and sends + returned string to <out> if specified. + + exception: + None, or a (type, value, traceback) tuple, e.g. from sys.exc_info(). If + None, we call sys.exc_info() and use its return value. + limit: + None or maximum number of stackframes to output. + out: + None or callable taking single <text> parameter or object with a + 'write' member that takes a single <text> parameter. + prefix: + Used to prefix all lines of text. + ''' + if exception is None: + exception = sys.exc_info() + etype, value, tb = exception + + if sys.version_info[0] == 2: + out2 = io.BytesIO() + else: + out2 = io.StringIO() + try: + + frames = [] + + # Get frames above point at which exception was caught - frames + # starting at top-level <module> or thread creation fn, and ending + # at the point in the catch: block from which we were called. + # + # These frames are not included explicitly in sys.exc_info()[2] and are + # also omitted by traceback.* functions, which makes for incomplete + # backtraces that miss much useful information. + # + for f in reversed(inspect.getouterframes(tb.tb_frame)): + ff = f[1], f[2], f[3], f[4][0].strip() + frames.append(ff) + + if 1: + # It's useful to see boundary between upper and lower frames. + frames.append( None) + + # Append frames from point in the try: block that caused the exception + # to be raised, to the point at which the exception was thrown. + # + # [One can get similar information using traceback.extract_tb(tb): + # for f in traceback.extract_tb(tb): + # frames.append(f) + # ] + for f in inspect.getinnerframes(tb): + ff = f[1], f[2], f[3], f[4][0].strip() + frames.append(ff) + + cwd = os.getcwd() + os.sep + if oneline: + if etype and value: + # The 'exception_text' variable below will usually be assigned + # something like '<ExceptionType>: <ExceptionValue>', unless + # there was no explanatory text provided (e.g. "raise Exception()"). + # In this case, str(value) will evaluate to ''. + exception_text = traceback.format_exception_only(etype, value)[0].strip() + filename, line, fnname, text = frames[-1] + if filename.startswith(cwd): + filename = filename[len(cwd):] + if not str(value): + # The exception doesn't have any useful explanatory text + # (for example, maybe it was raised by an expression like + # "assert <expression>" without a subsequent comma). In + # the absence of anything more helpful, print the code that + # raised the exception. + exception_text += ' (%s)' % text + line = '%s%s at %s:%s:%s()' % (prefix, exception_text, filename, line, fnname) + out2.write(line) + else: + out2.write( '%sBacktrace:\n' % prefix) + for frame in frames: + if frame is None: + out2.write( '%s ^except raise:\n' % prefix) + continue + filename, line, fnname, text = frame + if filename.startswith( cwd): + filename = filename[ len(cwd):] + if filename.startswith( './'): + filename = filename[ 2:] + out2.write( '%s %s:%s:%s(): %s\n' % ( + prefix, filename, line, fnname, text)) + + if etype and value: + out2.write( '%sException:\n' % prefix) + lines = traceback.format_exception_only( etype, value) + for line in lines: + out2.write( '%s %s' % ( prefix, line)) + + text = out2.getvalue() + + # Write text to <out> if specified. + out = getattr( out, 'write', out) + if callable( out): + out( text) + return text + + finally: + # clear things to avoid cycles. + exception = None + etype = None + value = None + tb = None + frames = None + + +def number_sep( s): + ''' + Simple number formatter, adds commas in-between thousands. <s> can + be a number or a string. Returns a string. + ''' + if not isinstance( s, str): + s = str( s) + c = s.find( '.') + if c==-1: c = len(s) + end = s.find('e') + if end == -1: end = s.find('E') + if end == -1: end = len(s) + ret = '' + for i in range( end): + ret += s[i] + if i<c-1 and (c-i-1)%3==0: + ret += ',' + elif i>c and i<end-1 and (i-c)%3==0: + ret += ',' + ret += s[end:] + return ret + +assert number_sep(1)=='1' +assert number_sep(12)=='12' +assert number_sep(123)=='123' +assert number_sep(1234)=='1,234' +assert number_sep(12345)=='12,345' +assert number_sep(123456)=='123,456' +assert number_sep(1234567)=='1,234,567' + + +class Stream: + ''' + Base layering abstraction for streams - abstraction for things like + sys.stdout to allow prefixing of all output, e.g. with a timestamp. + ''' + def __init__( self, stream): + self.stream = stream + def write( self, text): + self.stream.write( text) + +class StreamPrefix: + ''' + Prefixes output with a prefix, which can be a string or a callable that + takes no parameters and return a string. + ''' + def __init__( self, stream, prefix): + self.stream = stream + self.at_start = True + if callable(prefix): + self.prefix = prefix + else: + self.prefix = lambda : prefix + + def write( self, text): + if self.at_start: + text = self.prefix() + text + self.at_start = False + append_newline = False + if text.endswith( '\n'): + text = text[:-1] + self.at_start = True + append_newline = True + text = text.replace( '\n', '\n%s' % self.prefix()) + if append_newline: + text += '\n' + self.stream.write( text) + + def flush( self): + self.stream.flush() + + +def debug( text): + if callable(text): + text = text() + print( text) + +debug_periodic_t0 = [0] +def debug_periodic( text, override=0): + interval = 10 + t = time.time() + if t - debug_periodic_t0[0] > interval or override: + debug_periodic_t0[0] = t + debug(text) + + +def time_duration( seconds, verbose=False, s_format='%i'): + ''' + Returns string expressing an interval. + + seconds: + The duration in seconds + verbose: + If true, return like '4 days 1 hour 2 mins 23 secs', otherwise as + '4d3h2m23s'. + s_format: + If specified, use as printf-style format string for seconds. + ''' + x = abs(seconds) + ret = '' + i = 0 + for div, text in [ + ( 60, 'sec'), + ( 60, 'min'), + ( 24, 'hour'), + ( None, 'day'), + ]: + force = ( x == 0 and i == 0) + if div: + remainder = x % div + x = int( x/div) + else: + remainder = x + if not verbose: + text = text[0] + if remainder or force: + if verbose and remainder > 1: + # plural. + text += 's' + if verbose: + text = ' %s ' % text + if i == 0: + remainder = s_format % remainder + ret = '%s%s%s' % ( remainder, text, ret) + i += 1 + ret = ret.strip() + if ret == '': + ret = '0s' + if seconds < 0: + ret = '-%s' % ret + return ret + +assert time_duration( 303333) == '3d12h15m33s' +assert time_duration( 303333.33, s_format='%.1f') == '3d12h15m33.3s' +assert time_duration( 303333, verbose=True) == '3 days 12 hours 15 mins 33 secs' +assert time_duration( 303333.33, verbose=True, s_format='%.1f') == '3 days 12 hours 15 mins 33.3 secs' + +assert time_duration( 0) == '0s' +assert time_duration( 0, verbose=True) == '0 sec' + + +def date_time( t=None): + if t is None: + t = time.time() + return time.strftime( "%F-%T", time.gmtime( t)) + +def stream_prefix_time( stream): + ''' + Returns StreamPrefix that prefixes lines with time and elapsed time. + ''' + t_start = time.time() + def prefix_time(): + return '%s (+%s): ' % ( + time.strftime( '%T'), + time_duration( time.time() - t_start, s_format='0.1f'), + ) + return StreamPrefix( stream, prefix_time) + +def stdout_prefix_time(): + ''' + Changes sys.stdout to prefix time and elapsed time; returns original + sys.stdout. + ''' + ret = sys.stdout + sys.stdout = stream_prefix_time( sys.stdout) + return ret + + +def make_stream( out): + ''' + If <out> already has a .write() member, returns <out>. + + Otherwise a stream-like object with a .write() method that writes to <out>. + + out: + Where output is sent. + If None, output is lost. + Otherwise if an integer, we do: os.write( out, text) + Otherwise if callable, we do: out( text) + Otherwise we assume <out> is python stream or similar already. + ''' + if getattr( out, 'write', None): + return out + class Ret: + def flush(): + pass + ret = Ret() + if out is None: + ret.write = lambda text: None + elif isinstance( out, int): + ret.write = lambda text: os.write( out, text) + elif callable( out): + ret.write = out + else: + ret.write = lambda text: out.write( text) + return ret + + +def system_raw( + command, + out=None, + shell=True, + encoding='latin_1', + errors='strict', + buffer_len=-1, + ): + ''' + Runs command, writing output to <out> which can be an int fd, a python + stream or a Stream object. + + Args: + command: + The command to run. + out: + Where output is sent. + If None, output is lost. + If -1, output is sent to stdout and stderr. + Otherwise if an integer, we do: os.write( out, text) + Otherwise if callable, we do: out( text) + Otherwise we assume <out> is python stream or similar, and do: out.write(text) + shell: + Whether to run command inside a shell (see subprocess.Popen). + encoding: + Sepecify the encoding used to translate the command's output + to characters. + + Note that if <encoding> is None and we are being run by python3, + <out> will be passed bytes, not a string. + + Note that latin_1 will never raise a UnicodeDecodeError. + errors: + How to handle encoding errors; see docs for codecs module for + details. + buffer_len: + The number of bytes we attempt to read at a time. If -1 we read + output one line at a time. + + Returns: + subprocess's <returncode>, i.e. -N means killed by signal N, otherwise + the exit value (e.g. 12 if command terminated with exit(12)). + ''' + if out == -1: + stdin = 0 + stdout = 1 + stderr = 2 + else: + stdin = None + stdout = subprocess.PIPE + stderr = subprocess.STDOUT + child = subprocess.Popen( + command, + shell=shell, + stdin=stdin, + stdout=stdout, + stderr=stderr, + close_fds=True, + #encoding=encoding - only python-3.6+. + ) + + child_out = child.stdout + if encoding: + child_out = codecs.getreader( encoding)( child_out, errors) + + out = make_stream( out) + + if stdout == subprocess.PIPE: + if buffer_len == -1: + for line in child_out: + out.write( line) + else: + while 1: + text = child_out.read( buffer_len) + if not text: + break + out.write( text) + #decode( lambda : os.read( child_out.fileno(), 100), outfn, encoding) + + return child.wait() + +if __name__ == '__main__': + + if os.getenv( 'jtest_py_system_raw_test') == '1': + out = io.StringIO() + system_raw( + 'jtest_py_system_raw_test=2 python jlib.py', + sys.stdout, + encoding='utf-8', + #'latin_1', + errors='replace', + ) + print( repr( out.getvalue())) + + elif os.getenv( 'jtest_py_system_raw_test') == '2': + for i in range(256): + sys.stdout.write( chr(i)) + + +def system( + command, + verbose=None, + raise_errors=True, + out=None, + prefix=None, + rusage=False, + shell=True, + encoding=None, + errors='replace', + buffer_len=-1, + ): + ''' + Runs a command like os.system() or subprocess.*, but with more flexibility. + + We give control over where the command's output is sent, whether to return + the output and/or exit code, and whether to raise an exception if the + command fails. + + We also support the use of /usr/bin/time to gather rusage information. + + command: + The command to run. + verbose: + If true, we output information about the command that we run, and + its result. + + If callable or something with a .write() method, information is + sent to <verbose> itself. Otherwise it is sent to <out> (without + applying <prefix>). + raise_errors: + If true, we raise an exception if the command fails, otherwise we + return the failing error code or zero. + out: + Python stream, fd, callable or Stream instance to which output is + sent. + + If <out> is 'return', we buffer the output and return (e, + <output>). Note that if raise_errors is true, we only return if <e> + is zero. + + If -1, output is sent to stdout and stderr. + prefix: + If not None, should be prefix string or callable used to prefix + all output. [This is for convenience to avoid the need to do + out=StreamPrefix(...).] + rusage: + If true, we run via /usr/bin/time and return rusage string + containing information on execution. <raise_errors> and + out='return' are ignored. + shell: + Passed to underlying subprocess.Popen() call. + encoding: + Sepecify the encoding used to translate the command's output + to characters. Defaults to utf-8. + errors: + How to handle encoding errors; see docs for codecs module + for details. Defaults to 'replace' so we never raise a + UnicodeDecodeError. + buffer_len: + The number of bytes we attempt to read at a time. If -1 we read + output one line at a time. + + Returns: + If <rusage> is true, we return the rusage text. + + Else if raise_errors is true: + If the command failed, we raise an exception. + Else if <out> is 'return' we return the text output from the command. + Else we return None + + Else if <out> is 'return', we return (e, text) where <e> is the + command's exit code and <text> is the output from the command. + + Else we return <e>, the command's exit code. + ''' + if encoding is None: + if sys.version_info[0] == 2: + # python-2 doesn't seem to implement 'replace' properly. + encoding = None + errors = None + else: + encoding = 'utf-8' + errors = 'replace' + + out_original = out + if out is None: + out = sys.stdout + elif out == 'return': + # Store the output ourselves so we can return it. + out = io.StringIO() + else: + out = make_stream( out) + + if verbose: + if getattr( verbose, 'write', None): + pass + elif callable( verbose): + verbose = make_stream( verbose) + else: + verbose = out + + if prefix: + out = StreamPrefix( out, prefix) + + if verbose: + verbose.write( 'running: %s\n' % command) + + if rusage: + command2 = '' + command2 += '/usr/bin/time -o ubt-out -f "D=%D E=%D F=%F I=%I K=%K M=%M O=%O P=%P R=%r S=%S U=%U W=%W X=%X Z=%Z c=%c e=%e k=%k p=%p r=%r s=%s t=%t w=%w x=%x C=%C"' + command2 += ' ' + command2 += command + system_raw( command2, out, shell, encoding, errors, buffer_len=buffer_len) + with open('ubt-out') as f: + rusage_text = f.read() + #print 'have read rusage output: %r' % rusage_text + if rusage_text.startswith( 'Command '): + # Annoyingly, /usr/bin/time appears to write 'Command + # exited with ...' or 'Command terminated by ...' to the + # output file before the rusage info if command doesn't + # exit 0. + nl = rusage_text.find('\n') + rusage_text = rusage_text[ nl+1:] + return rusage_text + else: + e = system_raw( command, out, shell, encoding, errors, buffer_len=buffer_len) + + if verbose: + verbose.write( '[returned e=%s]\n' % e) + + if raise_errors: + if e: + raise Exception( 'command failed: %s' % command) + if out_original == 'return': + return out.getvalue() + return + + if out_original == 'return': + return e, out.getvalue() + return e + +def get_gitfiles( directory, submodules=False): + ''' + Returns list of all files known to git in <directory>; <directory> must be + somewhere within a git checkout. + + Returned names are all relative to <directory>. + + If .git directory, we also create <directory>/jtest-git-files. Otherwise we + assume a this file already exists. + ''' + if os.path.isdir( '%s/.git' % directory): + command = 'cd ' + directory + ' && git ls-files' + if submodules: + command += ' --recurse-submodules' + command += ' > jtest-git-files' + system( command, verbose=sys.stdout) + + with open( '%s/jtest-git-files' % directory, 'r') as f: + text = f.read() + ret = text.split( '\n') + return ret + +def get_git_id_raw( directory): + if not os.path.isdir( '%s/.git' % directory): + return + text = system( + f'cd {directory} && (PAGER= git show --pretty=oneline|head -n 1 && git diff)', + out='return', + ) + return text + +def get_git_id( directory, allow_none=False): + ''' + Returns text where first line is '<git-sha> <commit summary>' and remaining + lines contain output from 'git diff' in <directory>. + + directory: + Root of git checkout. + allow_none: + If true, we return None if <directory> is not a git checkout and + jtest-git-id file does not exist. + ''' + filename = f'{directory}/jtest-git-id' + text = get_git_id_raw( directory) + if text: + with open( filename, 'w') as f: + f.write( text) + elif os.path.isfile( filename): + with open( filename) as f: + text = f.read() + else: + if not allow_none: + raise Exception( f'Not in git checkout, and no file {filename}.') + text = None + return text + +class Args: + ''' + Iterates over argv items. Does getopt-style splitting of args starting with + single '-' character. + ''' + def __init__( self, argv): + self.argv = argv + self.pos = 0 + self.pos_sub = None + def next( self): + while 1: + if self.pos >= len(self.argv): + raise StopIteration() + arg = self.argv[self.pos] + if (not self.pos_sub + and arg.startswith('-') + and not arg.startswith('--') + ): + # Start splitting current arg. + self.pos_sub = 1 + if self.pos_sub and self.pos_sub >= len(arg): + # End of '-' sub-arg. + self.pos += 1 + self.pos_sub = None + continue + if self.pos_sub: + # Return '-' sub-arg. + ret = arg[self.pos_sub] + self.pos_sub += 1 + return f'-{ret}' + # Return normal arg. + self.pos += 1 + return arg + +def update_file( text, filename): + ''' + Writes <text> to <filename>. Does nothing if contents of <filename> are + already <text>. + ''' + try: + with open( filename) as f: + text0 = f.read() + except OSError: + text0 = None + if text == text0: + log( 'Unchanged: ' + filename) + else: + log( 'Updating: ' + filename) + # Write to temp file and rename, to ensure we are atomic. + filename_temp = f'{filename}-jlib-temp' + with open( filename_temp, 'w') as f: + f.write( text) + os.rename( filename_temp, filename) + + +def mtime( filename, default=0): + ''' + Returns mtime of file, or <default> if error - e.g. doesn't exist. + ''' + try: + return os.path.getmtime( filename) + except OSError: + return default + +def get_filenames( paths): + ''' + Yields each file in <paths>, walking any directories. + ''' + if isinstance( paths, str): + paths = (paths,) + for name in paths: + if os.path.isdir( name): + for dirpath, dirnames, filenames in os.walk( name): + for filename in filenames: + path = os.path.join( dirpath, filename) + yield path + else: + yield name + +def remove( path): + ''' + Removes file or directory, without raising exception if it doesn't exist. + + We assert-fail if the path still exists when we return, in case of + permission problems etc. + ''' + try: + os.remove( path) + except Exception: + pass + shutil.rmtree( path, ignore_errors=1) + assert not os.path.exists( path) + + +# Things for figuring out whether files need updating, using mtimes. +# +def newest( names): + ''' + Returns mtime of newest file in <filenames>. Returns 0 if no file exists. + ''' + assert isinstance( names, (list, tuple)) + assert names + ret_t = 0 + ret_name = None + for filename in get_filenames( names): + t = mtime( filename) + if t > ret_t: + ret_t = t + ret_name = filename + return ret_t, ret_name + +def oldest( names): + ''' + Returns mtime of oldest file in <filenames> or 0 if no file exists. + ''' + assert isinstance( names, (list, tuple)) + assert names + ret_t = None + ret_name = None + for filename in get_filenames( names): + t = mtime( filename) + if ret_t is None or t < ret_t: + ret_t = t + ret_name = filename + if ret_t is None: + ret_t = 0 + return ret_t, ret_name + +def update_needed( infiles, outfiles): + ''' + If any file in <infiles> is newer than any file in <outfiles>, returns + string description. Otherwise returns None. + ''' + in_tmax, in_tmax_name = newest( infiles) + out_tmin, out_tmin_name = oldest( outfiles) + if in_tmax > out_tmin: + text = f'{in_tmax_name} is newer than {out_tmin_name}' + return text + +def build( + infiles, + outfiles, + command, + force_rebuild=False, + out=None, + all_reasons=False, + verbose=True, + prefix=None, + ): + ''' + Ensures that <outfiles> are up to date using enhanced makefile-like + determinism of dependencies. + + Rebuilds <outfiles> by running <command> if we determine that any of them + are out of date. + + infiles: + Names of files that are read by <command>. Can be a single filename. If + an item is a directory, we expand to all filenames in the directory's + tree. + outfiles: + Names of files that are written by <command>. Can also be a single + filename. + command: + Command to run. + force_rebuild: + If true, we always re-run the command. + out: + A callable, passed to jlib.system(). If None, we use jlib.log() with + our caller's stack record. + all_reasons: + If true we check all ways for a build being needed, even if we already + know a build is needed; this only affects the diagnostic that we + output. + verbose: + Passed to jlib.system(). + prefix: + Passed to jlib.system(). + + We compare mtimes of <infiles> and <outfiles>, and we also detect changes + to the command itself. + + If any of infiles are newer than any of outfiles, or <command> is + different to contents of commandfile '<outfile[0]>.cmd, then truncates + commandfile and runs <command>. If <command> succeeds we writes <command> + to commandfile. + ''' + if isinstance( infiles, str): + infiles = (infiles,) + if isinstance( outfiles, str): + infiles = (outfiles,) + + if not out: + out_frame_record = inspect.stack()[1] + out = lambda text: log( text, nv=0, caller=out_frame_record) + + command_filename = f'{outfiles[0]}.cmd' + + reasons = [] + + if not reasons or all_reasons: + if force_rebuild: + reasons.append( 'force_rebuild was specified') + + if not reasons or all_reasons: + try: + with open( command_filename) as f: + command0 = f.read() + except Exception: + command0 = None + if command != command0: + if command0: + reasons.append( 'command has changed') + else: + reasons.append( 'no previous command') + + if not reasons or all_reasons: + reason = update_needed( infiles, outfiles) + if reason: + reasons.append( reason) + + if not reasons: + out( 'Already up to date: ' + ' '.join(outfiles)) + return + + if out: + out( 'Rebuilding because %s: %s' % ( + ', and '.join( reasons), + ' '.join(outfiles), + )) + + # Empty <command_filename) while we run the command so that if command + # fails but still creates target(s), then next time we will know target(s) + # are not up to date. + # + with open( command_filename, 'w') as f: + pass + + system( command, out=out, verbose=verbose, prefix=prefix) + + with open( command_filename, 'w') as f: + f.write( command) + + +def link_l_flags( sos): + ''' + Returns flags needed to link with items in <sos>. For each unique item we + use -L with parent directory, and -l with embedded name (without leading + 'lib' or trailing '.co'). + ''' + dirs = set() + names = [] + if isinstance( sos, str): + sos = (sos,) + for so in sos: + dir_ = os.path.dirname( so) + name = os.path.basename( so) + assert name.startswith( 'lib') + assert name.endswith ( '.so') + name = name[3:-3] + dirs.add( dir_) + names.append( name) + ret = '' + # Important to use sorted() here, otherwise ordering from set() is + # arbitrary causing occasional spurious rebuilds by jlib.build(). + for dir_ in sorted(dirs): + ret += f' -L {dir_}' + for name in names: + ret += f' -l {name}' + return ret |