Mercurial > hgrepos > Python2 > PyMuPDF
view mupdf-source/scripts/wrap/swig.py @ 29:f76e6575dca9 v1.26.4+1
+++++ v1.26.4+1
| author | Franz Glasner <fzglas.hg@dom66.de> |
|---|---|
| date | Fri, 19 Sep 2025 19:59:23 +0200 |
| parents | b50eed0cc0ef |
| children |
line wrap: on
line source
''' Support for using SWIG to generate language bindings from the C++ bindings. ''' import inspect import io import os import re import textwrap import jlib from . import cpp from . import csharp from . import rename from . import state from . import util def translate_ucdn_macros( build_dirs): ''' Returns string containing UCDN_* macros represented as enums. ''' out = io.StringIO() with open( f'{build_dirs.dir_mupdf}/include/mupdf/ucdn.h') as f: text = f.read() out.write( '\n') out.write( '\n') out.write( 'enum\n') out.write( '{\n') n = 0 for m in re.finditer('\n#define (UCDN_[A-Z0-9_]+) +([^\n]+)', text): out.write(f' {m.group(1)} = {m.group(2)},\n') n += 1 out.write( '};\n') out.write( '\n') assert n return out.getvalue() def _csharp_unicode_prefix(): ''' Returns typemaps that automatically convert C# strings (which are utf16) into utf8 when calling MuPDF, and convert strings returned by MuPDF (which are utf8) into utf16. We return empty string if not on Windows, because Mono appears to already work. ''' if not state.state_.windows: # Mono on Linux already seems to use utf8. return '' text = textwrap.dedent(''' // This ensures that our code below overrides whatever is defined // in std_string.i and any later `%include "std_string.i"` is // ignored. %include "std_string.i" // See https://github.com/swig/swig/pull/2364. We also add typemaps // for `const char*`. %{ #include <string> %} namespace std { %typemap(imtype, inattributes="[global::System.Runtime.InteropServices.MarshalAs(global::System.Runtime.InteropServices.UnmanagedType.LPUTF8Str)]", outattributes="[return: global::System.Runtime.InteropServices.MarshalAs(global::System.Runtime.InteropServices.UnmanagedType.LPUTF8Str)]", directorinattributes="[global::System.Runtime.InteropServices.MarshalAs(global::System.Runtime.InteropServices.UnmanagedType.LPUTF8Str)]", directoroutattributes="[return: global::System.Runtime.InteropServices.MarshalAs(global::System.Runtime.InteropServices.UnmanagedType.LPUTF8Str)]" ) string "string" %typemap(imtype, inattributes="[global::System.Runtime.InteropServices.MarshalAs(global::System.Runtime.InteropServices.UnmanagedType.LPUTF8Str)]", outattributes="[return: global::System.Runtime.InteropServices.MarshalAs(global::System.Runtime.InteropServices.UnmanagedType.LPUTF8Str)]", directorinattributes="[global::System.Runtime.InteropServices.MarshalAs(global::System.Runtime.InteropServices.UnmanagedType.LPUTF8Str)]", directoroutattributes="[return: global::System.Runtime.InteropServices.MarshalAs(global::System.Runtime.InteropServices.UnmanagedType.LPUTF8Str)]" ) const string & "string" %typemap(imtype, inattributes="[global::System.Runtime.InteropServices.MarshalAs(global::System.Runtime.InteropServices.UnmanagedType.LPUTF8Str)]", outattributes="[return: global::System.Runtime.InteropServices.MarshalAs(global::System.Runtime.InteropServices.UnmanagedType.LPUTF8Str)]", directorinattributes="[global::System.Runtime.InteropServices.MarshalAs(global::System.Runtime.InteropServices.UnmanagedType.LPUTF8Str)]", directoroutattributes="[return: global::System.Runtime.InteropServices.MarshalAs(global::System.Runtime.InteropServices.UnmanagedType.LPUTF8Str)]" ) const char* "string" } ''') return text def build_swig( state_: state.State, build_dirs: state.BuildDirs, generated, language='python', swig_command='swig', check_regress=False, force_rebuild=False, ): ''' Builds python or C# wrappers for all mupdf_* functions and classes, by creating a .i file that #include's our generated C++ header files and running swig. build_dirs A BuildDirs instance. generated. A Generated instance. language The output language, must be 'python' or 'csharp'. swig Location of swig binary. check_regress If true, we fail with error if generated .i file already exists and differs from our new content. ''' assert isinstance( state_, state.State) assert isinstance(build_dirs, state.BuildDirs), type(build_dirs) assert isinstance(generated, cpp.Generated), type(generated) assert language in ('python', 'csharp') # Find version of swig. (We use quotes around <swig> to make things work on # Windows.) e, swig_location = jlib.system( f'which "{swig_command}"', raise_errors=0, out='return', verbose=0) if e == 0: jlib.log(f'{swig_location=}') t = jlib.system( f'"{swig_command}" -version', out='return', verbose=0) jlib.log1('SWIG version info:\n========\n{t}\n========') m = re.search( 'SWIG Version ([0-9]+)[.]([0-9]+)[.]([0-9]+)', t) assert m swig_major = int( m.group(1)) jlib.log(f'{m.group()}') # Create a .i file for SWIG. # common = textwrap.dedent(f''' #include <stdexcept> #include "mupdf/functions.h" #include "mupdf/classes.h" #include "mupdf/classes2.h" #include "mupdf/internal.h" #include "mupdf/exceptions.h" #include "mupdf/extra.h" #ifdef NDEBUG static bool g_mupdf_trace_director = false; static bool g_mupdf_trace_exceptions = false; #else static bool g_mupdf_trace_director = mupdf::internal_env_flag("MUPDF_trace_director"); static bool g_mupdf_trace_exceptions = mupdf::internal_env_flag("MUPDF_trace_exceptions"); #endif ''' ) if language == 'csharp': common += textwrap.dedent(f''' /* This is required otherwise compiling the resulting C++ code fails with: error: use of undeclared identifier 'SWIG_fail' But no idea whether it is the 'correct' thing to do; seems odd that SWIG doesn't define SWIG_fail itself. */ #define SWIG_fail throw std::runtime_error( e.what()); ''') if language == 'python': common += textwrap.dedent(f''' static std::string to_stdstring(PyObject* s) {{ PyObject* repr_str = PyUnicode_AsEncodedString(s, "utf-8", "~E~"); #ifdef Py_LIMITED_API const char* repr_str_s = PyBytes_AsString(repr_str); #else const char* repr_str_s = PyBytes_AS_STRING(repr_str); #endif std::string ret = repr_str_s; Py_DECREF(repr_str); Py_DECREF(s); return ret; }} static std::string py_repr(PyObject* x) {{ if (!x) return "<C_nullptr>"; PyObject* s = PyObject_Repr(x); return to_stdstring(s); }} static std::string py_str(PyObject* x) {{ if (!x) return "<C_nullptr>"; PyObject* s = PyObject_Str(x); return to_stdstring(s); }} /* Returns a Python `bytes` containing a copy of a `fz_buffer`'s data. If <clear> is true we also clear and trim the buffer. */ PyObject* ll_fz_buffer_to_bytes_internal(fz_buffer* buffer, int clear) {{ unsigned char* c = NULL; size_t len = {rename.namespace_ll_fn('fz_buffer_storage')}(buffer, &c); PyObject* ret = PyBytes_FromStringAndSize((const char*) c, (Py_ssize_t) len); if (clear) {{ /* We mimic the affects of fz_buffer_extract(), which leaves the buffer with zero capacity. */ {rename.namespace_ll_fn('fz_clear_buffer')}(buffer); {rename.namespace_ll_fn('fz_trim_buffer')}(buffer); }} return ret; }} /* Returns a Python `memoryview` for specified memory. */ PyObject* python_memoryview_from_memory( void* data, size_t size, int writable) {{ return PyMemoryView_FromMemory( (char*) data, (Py_ssize_t) size, writable ? PyBUF_WRITE : PyBUF_READ ); }} /* Returns a Python `memoryview` for a `fz_buffer`'s data. */ PyObject* ll_fz_buffer_storage_memoryview(fz_buffer* buffer, int writable) {{ unsigned char* data = NULL; size_t len = {rename.namespace_ll_fn('fz_buffer_storage')}(buffer, &data); return python_memoryview_from_memory( data, len, writable); }} /* Creates Python bytes from copy of raw data. */ PyObject* raw_to_python_bytes(const unsigned char* c, size_t len) {{ return PyBytes_FromStringAndSize((const char*) c, (Py_ssize_t) len); }} /* Creates Python bytes from copy of raw data. */ PyObject* raw_to_python_bytes(const void* c, size_t len) {{ return PyBytes_FromStringAndSize((const char*) c, (Py_ssize_t) len); }} /* The SWIG wrapper for this function returns a SWIG proxy for a 'const unsigned char*' pointing to the raw data of a python bytes. This proxy can then be passed from Python to functions that take a 'const unsigned char*'. For example to create a MuPDF fz_buffer* from a copy of a Python bytes instance: bs = b'qwerty' buffer_ = mupdf.fz_new_buffer_from_copied_data(mupdf.python_buffer_data(bs), len(bs)) */ const unsigned char* python_buffer_data( const unsigned char* PYTHON_BUFFER_DATA, size_t PYTHON_BUFFER_SIZE ) {{ return PYTHON_BUFFER_DATA; }} unsigned char* python_mutable_buffer_data( unsigned char* PYTHON_BUFFER_MUTABLE_DATA, size_t PYTHON_BUFFER_MUTABLE_SIZE ) {{ return PYTHON_BUFFER_MUTABLE_DATA; }} /* Casts an integer to a pdf_obj*. Used to convert SWIG's int values for PDF_ENUM_NAME_* into {rename.class_('pdf_obj')}'s. */ pdf_obj* obj_enum_to_obj(int n) {{ return (pdf_obj*) (intptr_t) n; }} /* SWIG-friendly alternative to {rename.ll_fn('pdf_set_annot_color')}(). */ void {rename.ll_fn('pdf_set_annot_color2')}(pdf_annot *annot, int n, float color0, float color1, float color2, float color3) {{ float color[] = {{ color0, color1, color2, color3 }}; return {rename.namespace_ll_fn('pdf_set_annot_color')}(annot, n, color); }} /* SWIG-friendly alternative to {rename.ll_fn('pdf_set_annot_interior_color')}(). */ void {rename.ll_fn('pdf_set_annot_interior_color2')}(pdf_annot *annot, int n, float color0, float color1, float color2, float color3) {{ float color[] = {{ color0, color1, color2, color3 }}; return {rename.namespace_ll_fn('pdf_set_annot_interior_color')}(annot, n, color); }} /* SWIG-friendly alternative to `fz_fill_text()`. */ void ll_fz_fill_text2( fz_device* dev, const fz_text* text, fz_matrix ctm, fz_colorspace* colorspace, float color0, float color1, float color2, float color3, float alpha, fz_color_params color_params ) {{ float color[] = {{color0, color1, color2, color3}}; return {rename.namespace_ll_fn( 'fz_fill_text')}(dev, text, ctm, colorspace, color, alpha, color_params); }} std::vector<unsigned char> {rename.fn('fz_memrnd2')}(int length) {{ std::vector<unsigned char> ret(length); {rename.namespace_fn('fz_memrnd')}(&ret[0], length); return ret; }} /* mupdfpy optimisation for copying raw data into pixmap. `samples` must have enough data to fill the pixmap. */ void ll_fz_pixmap_copy_raw( fz_pixmap* pm, const void* samples) {{ memcpy(pm->samples, samples, pm->stride * pm->h); }} ''') common += textwrap.dedent(f''' /* SWIG-friendly alternative to fz_runetochar(). */ std::vector<unsigned char> {rename.fn('fz_runetochar2')}(int rune) {{ std::vector<unsigned char> buffer(10); int n = {rename.namespace_ll_fn('fz_runetochar')}((char*) &buffer[0], rune); assert(n < sizeof(buffer)); buffer.resize(n); return buffer; }} /* SWIG-friendly alternatives to fz_make_bookmark() and {rename.fn('fz_lookup_bookmark')}(), using long long instead of fz_bookmark because SWIG appears to treat fz_bookmark as an int despite it being a typedef for intptr_t, so ends up slicing. */ long long unsigned {rename.ll_fn('fz_make_bookmark2')}(fz_document* doc, fz_location loc) {{ fz_bookmark bm = {rename.namespace_ll_fn('fz_make_bookmark')}(doc, loc); return (long long unsigned) bm; }} fz_location {rename.ll_fn('fz_lookup_bookmark2')}(fz_document *doc, long long unsigned mark) {{ return {rename.namespace_ll_fn('fz_lookup_bookmark')}(doc, (fz_bookmark) mark); }} {rename.namespace_class('fz_location')} {rename.fn('fz_lookup_bookmark2')}( {rename.namespace_class('fz_document')} doc, long long unsigned mark) {{ return {rename.namespace_class('fz_location')}( {rename.ll_fn('fz_lookup_bookmark2')}(doc.m_internal, mark)); }} struct {rename.fn('fz_convert_color2_v')} {{ float v0; float v1; float v2; float v3; }}; /* SWIG-friendly alternative for {rename.ll_fn('fz_convert_color')}(), taking `float* sv`. */ void {rename.ll_fn('fz_convert_color2')}( fz_colorspace *ss, float* sv, fz_colorspace *ds, {rename.fn('fz_convert_color2_v')}* dv, fz_colorspace *is, fz_color_params params ) {{ //float sv[] = {{ sv0, sv1, sv2, sv3}}; {rename.namespace_ll_fn('fz_convert_color')}(ss, sv, ds, &dv->v0, is, params); }} /* SWIG-friendly alternative for {rename.ll_fn('fz_convert_color')}(), taking four explicit `float` values for `sv`. */ void {rename.ll_fn('fz_convert_color2')}( fz_colorspace *ss, float sv0, float sv1, float sv2, float sv3, fz_colorspace *ds, {rename.fn('fz_convert_color2_v')}* dv, fz_colorspace *is, fz_color_params params ) {{ float sv[] = {{ sv0, sv1, sv2, sv3}}; {rename.namespace_ll_fn('fz_convert_color')}(ss, sv, ds, &dv->v0, is, params); }} /* SWIG- Director class to allow fz_set_warning_callback() and fz_set_error_callback() to be used with Python callbacks. Note that we rename print() to _print() to match what SWIG does. */ struct DiagnosticCallback {{ /* `description` must be "error" or "warning". */ DiagnosticCallback(const char* description) : m_description(description) {{ #ifndef NDEBUG if (g_mupdf_trace_director) {{ std::cerr << __FILE__ << ":" << __LINE__ << ":" << __FUNCTION__ << ":" << " DiagnosticCallback[" << m_description << "]() constructor." << "\\n"; }} #endif if (m_description == "warning") {{ mupdf::ll_fz_set_warning_callback( s_print, this); }} else if (m_description == "error") {{ mupdf::ll_fz_set_error_callback( s_print, this); }} else {{ std::cerr << __FILE__ << ":" << __LINE__ << ":" << __FUNCTION__ << ":" << " DiagnosticCallback() constructor" << " Unrecognised description: " << m_description << "\\n"; assert(0); }} }} virtual void _print( const char* message) {{ #ifndef NDEBUG if (g_mupdf_trace_director) {{ std::cerr << __FILE__ << ":" << __LINE__ << ":" << __FUNCTION__ << ":" << " DiagnosticCallback[" << m_description << "]::_print()" << " called (no derived class?)" << " message: " << message << "\\n"; }} #endif }} virtual ~DiagnosticCallback() {{ #ifndef NDEBUG if (g_mupdf_trace_director) {{ std::cerr << __FILE__ << ":" << __LINE__ << ":" << __FUNCTION__ << ":" << " ~DiagnosticCallback[" << m_description << "]() destructor called" << " this=" << this << "\\n"; }} #endif }} static void s_print( void* self0, const char* message) {{ DiagnosticCallback* self = (DiagnosticCallback*) self0; try {{ return self->_print( message); }} catch (std::exception& e) {{ /* It's important to swallow any exception from self->_print() because fz_set_warning_callback() and fz_set_error_callback() specifically require that the callback does not throw. But we always output a diagnostic. */ std::cerr << "DiagnosticCallback[" << self->m_description << "]::s_print()" << " ignoring exception from _print(): " << e.what() << "\\n"; }} }} std::string m_description; }}; struct StoryPositionsCallback {{ StoryPositionsCallback() {{ //printf( "StoryPositionsCallback() constructor\\n"); }} virtual void call( const fz_story_element_position* position) = 0; static void s_call( fz_context* ctx, void* self0, const fz_story_element_position* position) {{ //printf( "StoryPositionsCallback::s_call()\\n"); (void) ctx; StoryPositionsCallback* self = (StoryPositionsCallback*) self0; self->call( position); }} virtual ~StoryPositionsCallback() {{ //printf( "StoryPositionsCallback() destructor\\n"); }} }}; void ll_fz_story_positions_director( fz_story *story, StoryPositionsCallback* cb) {{ //printf( "ll_fz_story_positions_director()\\n"); {rename.namespace_ll_fn('fz_story_positions')}( story, StoryPositionsCallback::s_call, cb ); }} void Pixmap_set_alpha_helper( int balen, int n, int data_len, int zero_out, unsigned char* data, fz_pixmap* pix, int premultiply, int bground, const std::vector<int>& colors, const std::vector<int>& bgcolor ) {{ int i = 0; int j = 0; int k = 0; int data_fix = 255; while (i < balen) {{ unsigned char alpha = data[k]; if (zero_out) {{ for (j = i; j < i+n; j++) {{ if (pix->samples[j] != (unsigned char) colors[j - i]) {{ data_fix = 255; break; }} else {{ data_fix = 0; }} }} }} if (data_len) {{ if (data_fix == 0) {{ pix->samples[i+n] = 0; }} else {{ pix->samples[i+n] = alpha; }} if (premultiply && !bground) {{ for (j = i; j < i+n; j++) {{ pix->samples[j] = fz_mul255(pix->samples[j], alpha); }} }} else if (bground) {{ for (j = i; j < i+n; j++) {{ int m = (unsigned char) bgcolor[j - i]; pix->samples[j] = m + fz_mul255((pix->samples[j] - m), alpha); }} }} }} else {{ pix->samples[i+n] = data_fix; }} i += n+1; k += 1; }} }} void page_merge_helper( {rename.namespace_class('pdf_obj')}& old_annots, {rename.namespace_class('pdf_graft_map')}& graft_map, {rename.namespace_class('pdf_document')}& doc_des, {rename.namespace_class('pdf_obj')}& new_annots, int n ) {{ #define PDF_NAME2(X) {rename.namespace_class('pdf_obj')}(PDF_NAME(X)) for ( int i=0; i<n; ++i) {{ {rename.namespace_class('pdf_obj')} o = {rename.namespace_fn('pdf_array_get')}( old_annots, i); if ({rename.namespace_fn('pdf_dict_gets')}( o, "IRT").m_internal) continue; {rename.namespace_class('pdf_obj')} subtype = {rename.namespace_fn('pdf_dict_get')}( o, PDF_NAME2(Subtype)); if ( {rename.namespace_fn('pdf_name_eq')}( subtype, PDF_NAME2(Link))) continue; if ( {rename.namespace_fn('pdf_name_eq')}( subtype, PDF_NAME2(Popup))) continue; if ( {rename.namespace_fn('pdf_name_eq')}( subtype, PDF_NAME2(Widget))) {{ /* fixme: C++ API doesn't yet wrap fz_warn() - it excludes all variadic fns. */ //mupdf::fz_warn( "skipping widget annotation"); continue; }} {rename.namespace_fn('pdf_dict_del')}( o, PDF_NAME2(Popup)); {rename.namespace_fn('pdf_dict_del')}( o, PDF_NAME2(P)); {rename.namespace_class('pdf_obj')} copy_o = {rename.namespace_fn('pdf_graft_mapped_object')}( graft_map, o); {rename.namespace_class('pdf_obj')} annot = {rename.namespace_fn('pdf_new_indirect')}( doc_des, {rename.namespace_fn('pdf_to_num')}( copy_o), 0); {rename.namespace_fn('pdf_array_push')}( new_annots, annot); }} #undef PDF_NAME2 }} ''') common += generated.swig_cpp common += translate_ucdn_macros( build_dirs) text = '' text += '%module(directors="1") mupdf\n' jlib.log(f'{build_dirs.Py_LIMITED_API=}') text += f'%begin %{{\n' if build_dirs.Py_LIMITED_API: # e.g. 0x03080000 text += textwrap.dedent(f''' /* Use Python Stable ABI with earliest Python version that we support. */ #define Py_LIMITED_API {build_dirs.Py_LIMITED_API} /* These seem to be mistakenly undefined when Py_LIMITED_API is defined, so we force the values from Python.h. Also see https://github.com/python/cpython/issues/98680. */ #ifndef PyBUF_READ #define PyBUF_READ 0x100 #endif #ifndef PyBUF_WRITE #define PyBUF_WRITE 0x200 #endif ''') text += textwrap.dedent(f''' /* This seems to be necessary on some Windows machines with Py_LIMITED_API, otherwise compilation can fail because free() and malloc() are not declared. */ #include <stdlib.h> ''') text += f'%}}\n' # https://www.mono-project.com/docs/advanced/pinvoke/ # # > Mono on all platforms currently uses UTF-8 encoding for all string # > marshaling operations. # if language == 'csharp': text += _csharp_unicode_prefix() for i in generated.virtual_fnptrs: text += f'%feature("director") {i};\n' text += f'%feature("director") DiagnosticCallback;\n' text += f'%feature("director") StoryPositionsCallback;\n' text += textwrap.dedent( ''' %feature("director:except") { if ($error != NULL) { /* This is how we can end up here: 1. Python code calls a function in the Python `mupdf` module. 2. - which calls SWIG C++ code. 3. - which calls MuPDF C++ API wrapper function. 4. - which calls MuPDF C code which calls an MuPDF struct's function pointer. 5. - which calls MuPDF C++ API Director wrapper (e.g. mupdf::FzDevice2) virtual function. 6. - which calls SWIG Director C++ code. 7. - which calls Python derived class's method, which raises a Python exception. The exception propagates back up the above stack, being converted into different exception representations as it goes: 6. SWIG Director C++ code (here). We raise a C++ exception. 5. MuPDF C++ API Director wrapper converts the C++ exception into a MuPDF fz_try/catch exception. 4. MuPDF C code allows the exception to propagate or catches and rethrows or throws a new fz_try/catch exception. 3. MuPDF C++ API wrapper function converts the fz_try/catch exception into a C++ exception. 2. SWIG C++ code converts the C++ exception into a Python exception. 1. Python code receives the Python exception. So the exception changes from a Python exception, to a C++ exception, to a fz_try/catch exception, to a C++ exception, and finally back into a Python exception. Each of these stages is necessary. In particular we cannot let the first C++ exception propagate directly through MuPDF C code without being a fz_try/catch exception, because it would mess up MuPDF C code's fz_try/catch exception stack. Unfortuntately MuPDF fz_try/catch exception strings are limited to 256 characters so some or all of our detailed backtrace information is lost. */ /* Get text description of the Python exception. */ PyObject* etype; PyObject* obj; PyObject* trace; PyErr_Fetch( &etype, &obj, &trace); /* Looks like PyErr_GetExcInfo() fails here, returning NULL.*/ /* PyErr_GetExcInfo( &etype, &obj, &trace); std::cerr << "PyErr_GetExcInfo(): etype: " << py_str(etype) << "\\n"; std::cerr << "PyErr_GetExcInfo(): obj: " << py_str(obj) << "\\n"; std::cerr << "PyErr_GetExcInfo(): trace: " << py_str(trace) << "\\n"; */ std::string message = "Director error: " + py_str(etype) + ": " + py_str(obj) + "\\n"; if (g_mupdf_trace_director) { /* __FILE__ and __LINE__ are not useful here because SWIG makes them point to the generic .i code. */ std::cerr << "========\\n"; std::cerr << "g_mupdf_trace_director set: Converting Python error into C++ exception:" << "\\n"; #ifndef _WIN32 std::cerr << " function: " << __PRETTY_FUNCTION__ << "\\n"; #endif std::cerr << " etype: " << py_str(etype) << "\\n"; std::cerr << " obj: " << py_str(obj) << "\\n"; std::cerr << " trace: " << py_str(trace) << "\\n"; std::cerr << "========\\n"; } PyObject* traceback = PyImport_ImportModule("traceback"); if (traceback) { /* Use traceback.format_tb() to get backtrace. */ if (0) { message += "Traceback (from traceback.format_tb()):\\n"; PyObject* traceback_dict = PyModule_GetDict(traceback); PyObject* format_tb = PyDict_GetItem(traceback_dict, PyString_FromString("format_tb")); PyObject* ret = PyObject_CallFunctionObjArgs(format_tb, trace, NULL); PyObject* iter = PyObject_GetIter(ret); for(;;) { PyObject* item = PyIter_Next(iter); if (!item) break; message += py_str(item); Py_DECREF(item); } /* `format_tb` and `traceback_dict` are borrowed references. */ Py_XDECREF(iter); Py_XDECREF(ret); Py_XDECREF(traceback); } /* Use exception_info() (copied from mupdf/scripts/jlib.py) to get detailed backtrace. */ if (1) { PyObject* globals = PyEval_GetGlobals(); PyObject* exception_info = PyDict_GetItemString(globals, "exception_info"); PyObject* string_return = PyUnicode_FromString("return"); PyObject* ret = PyObject_CallFunctionObjArgs( exception_info, trace, Py_None, string_return, NULL ); Py_XDECREF(string_return); message += py_str(ret); Py_XDECREF(ret); } } else { message += "[No backtrace available.]\\n"; } Py_XDECREF(etype); Py_XDECREF(obj); Py_XDECREF(trace); message += "Exception was from C++/Python callback:\\n"; message += " "; #ifdef _WIN32 message += __FUNCTION__; #else message += __PRETTY_FUNCTION__; #endif message += "\\n"; if (1 || g_mupdf_trace_director) { std::cerr << "========\\n"; std::cerr << "Director exception handler, message is:\\n" << message << "\\n"; std::cerr << "========\\n"; } /* SWIG 4.1 documentation talks about throwing a Swig::DirectorMethodException here, but this doesn't work for us because it sets Python's error state again, which makes the next SWIG call of a C/C++ function appear to fail. //throw Swig::DirectorMethodException(); */ throw std::runtime_error( message.c_str()); } } ''') # Ignore all C MuPDF functions; SWIG will still look at the C++ API in # namespace mudf. for fnname in generated.c_functions: if fnname in ( 'pdf_annot_type', 'pdf_widget_type', 'pdf_zugferd_profile', ): # These are also enums which we don't want to ignore. SWIGing the # functions is hopefully harmless. pass elif fnname in ('x0', 'y0', 'x1', 'y1'): # Windows appears to have functions called y0() and y1() e.g. in: # # C:\\Program Files (x86)\\Windows Kits\\10\\Include\\10.0.18362.0\\ucrt\\corecrt_math.h # # If we use `%ignore` with these, e.g. `%ignore ::y0`, swig # unhelpfully seems to also ignore any member variables called `y0` # or `y1`. # jlib.log('Not ignoring {fnname=} because breaks wrapping of fz_rect.') pass else: text += f'%ignore ::{fnname};\n' # Attempt to move C structs out of the way to allow wrapper classes to have # the same name as the struct they wrap. Unfortunately this causes a small # number of obscure errors from SWIG. if 0: for name in generated.c_structs: text += f'%rename(lll_{name}) ::{name};\n' for i in ( 'fz_append_vprintf', 'fz_error_stack_slot', 'fz_format_string', 'fz_vsnprintf', 'fz_vthrow', 'fz_vwarn', 'fz_write_vprintf', 'fz_vlog_error_printf', 'fz_utf8_from_wchar', 'fz_wchar_from_utf8', 'fz_fopen_utf8', 'fz_remove_utf8', 'fz_argv_from_wargv', 'fz_free_argv', 'fz_stdods', ): text += f'%ignore {i};\n' text += f'%ignore {rename.method(None, i)};\n' text += textwrap.dedent(f''' // Not implemented in mupdf.so: fz_colorspace_name_process_colorants %ignore fz_colorspace_name_process_colorants; %ignore fz_argv_from_wargv; %ignore fz_open_file_w; %ignore {rename.ll_fn('fz_append_vprintf')}; %ignore {rename.ll_fn('fz_error_stack_slot_s')}; %ignore {rename.ll_fn('fz_format_string')}; %ignore {rename.ll_fn('fz_vsnprintf')}; %ignore {rename.ll_fn('fz_vthrow')}; %ignore {rename.ll_fn('fz_vwarn')}; %ignore {rename.ll_fn('fz_write_vprintf')}; %ignore {rename.ll_fn('fz_vlog_error_printf')}; %ignore {rename.ll_fn('fz_open_file_w')}; // Ignore custom C++ variadic fns. %ignore {rename.ll_fn('pdf_dict_getlv')}; %ignore {rename.ll_fn('pdf_dict_getl')}; %ignore {rename.fn('pdf_dict_getlv')}; %ignore {rename.fn('pdf_dict_getl')}; // SWIG can't handle this because it uses a valist. %ignore {rename.ll_fn('Memento_vasprintf')}; %ignore {rename.fn('Memento_vasprintf')}; // These appear to be not present in Windows debug builds. %ignore fz_assert_lock_held; %ignore fz_assert_lock_not_held; %ignore fz_lock_debug_lock; %ignore fz_lock_debug_unlock; %ignore Memento_cpp_new; %ignore Memento_cpp_delete; %ignore Memento_cpp_new_array; %ignore Memento_cpp_delete_array; %ignore Memento_showHash; // asprintf() isn't available on Windows, so exclude Memento_asprintf because // it is #define-d to asprintf. %ignore {rename.ll_fn('Memento_asprintf')}; %ignore {rename.fn('Memento_asprintf')}; // Might prefer to #include mupdf/exceptions.h and make the // %exception block below handle all the different exception types, // but swig-3 cannot parse 'throw()' in mupdf/exceptions.h. // // So for now we just #include <stdexcept> and handle // std::exception only. %include "typemaps.i" %include "cpointer.i" // This appears to allow python to call fns taking an int64_t. %include "stdint.i" /* This is only documented for Ruby, but is mentioned for Python at https://sourceforge.net/p/swig/mailman/message/4867286/. It makes the Python wrapper for `FzErrorBase` inherit Python's `Exception` instead of `object`, which in turn means it can be caught in Python with `except Exception as e: ...` or similar. Note that while it will have the underlying C++ class's `what()` method, this is not used by the `__str__()` and `__repr__()` methods. Instead: `__str__()` appears to return a tuple of the constructor args that were originally used to create the exception object with `PyObject_CallObject(class_, args)`. `__repr__()` returns a SWIG-style string such as `<texcept.MyError; proxy of <Swig Object of type 'MyError *' at 0xb61ebfabc00> >`. We explicitly overwrite `__str__()` to call `what()`. */ %feature("exceptionclass") FzErrorBase; %{{ ''') text += common text += textwrap.dedent(f''' %}} %include exception.i %include std_string.i %include carrays.i %include cdata.i %include std_vector.i %include std_map.i {"%include argcargv.i" if language=="python" else ""} %array_class(unsigned char, uchar_array); %include <cstring.i> namespace std {{ %template(vectoruc) vector<unsigned char>; %template(vectori) vector<int>; %template(vectorf) vector<float>; %template(vectord) vector<double>; %template(vectors) vector<std::string>; %template(map_string_int) map<std::string, int>; %template(vectorq) vector<{rename.namespace_class("fz_quad")}>; %template(vector_search_page2_hit) vector<fz_search_page2_hit>; %template(vector_fz_font_ucs_gid) vector<fz_font_ucs_gid>; %template(vector_fz_point) vector<fz_point>; }}; // Make sure that operator++() gets converted to __next__(). // // Note that swig already seems to do: // // operator* => __ref__ // operator== => __eq__ // operator!= => __ne__ // operator-> => __deref__ // // Just need to add this method to containers that already have // begin() and end(): // def __iter__( self): // return CppIterator( self) // %rename(__increment__) *::operator++; // Create fns that give access to arrays of some basic types, e.g. bytes_getitem(). // %array_functions(unsigned char, bytes); // Useful for fz_stroke_state::dash_list[]. %array_functions(float, floats); ''') if language == 'python': text += generated.swig_python_exceptions.getvalue() text += textwrap.dedent(f''' // Ensure SWIG handles OUTPUT params. // %include "cpointer.i" ''') if swig_major < 4: text += textwrap.dedent(f''' // SWIG version is less than 4 so swig is not able to copy // across comments from header file into generated code. The // next best thing is to use autodoc to make swig at least show // some generic information about arg types. // %feature("autodoc", "3"); ''') text += textwrap.dedent(f''' // Tell swig about pdf_clean_file()'s (int,argv)-style args: %apply (int ARGC, char **ARGV) {{ (int retainlen, char *retainlist[]) }} ''') if language == 'python': text += textwrap.dedent( ''' %include pybuffer.i /* Convert Python buffer to (const unsigned char*, size_t) pair for python_buffer_data(). */ %pybuffer_binary( const unsigned char* PYTHON_BUFFER_DATA, size_t PYTHON_BUFFER_SIZE ); /* Convert Python buffer to (unsigned char*, size_t) pair for python_mutable_bytes_data(). */ %pybuffer_mutable_binary( unsigned char* PYTHON_BUFFER_MUTABLE_DATA, size_t PYTHON_BUFFER_MUTABLE_SIZE ); ''' ) text += common if language == 'python': text += textwrap.dedent(f''' %pointer_functions(int, pint); %pythoncode %{{ import inspect import os import re import sys import traceback def log( text): print( text, file=sys.stderr) g_mupdf_trace_director = (os.environ.get('MUPDF_trace_director') == '1') def fz_lookup_metadata(document, key): """ Like fz_lookup_metadata2() but returns None on error instead of raising exception. """ try: return fz_lookup_metadata2(document, key) except Exception: return {rename.class_('fz_document')}.{rename.method('fz_document', 'fz_lookup_metadata')} \ = fz_lookup_metadata def pdf_lookup_metadata(document, key): """ Likepsd_lookup_metadata2() but returns None on error instead of raising exception. """ try: return pdf_lookup_metadata2(document, key) except Exception: return {rename.class_('pdf_document')}.{rename.method('pdf_document', 'pdf_lookup_metadata')} \ = pdf_lookup_metadata ''') exception_info_text = inspect.getsource(jlib.exception_info) text += 'import inspect\n' text += 'import io\n' text += 'import os\n' text += 'import sys\n' text += 'import traceback\n' text += 'import types\n' text += exception_info_text if language == 'python': # Make some additions to the generated Python module. # # E.g. python wrappers for functions that take out-params should return # tuples. # text += generated.swig_python text += generated.swig_python_set_error_classes.getvalue() def set_class_method(struct, fn): return f'{rename.class_(struct)}.{rename.method(struct, fn)} = {fn}' text += textwrap.dedent(f''' # Wrap fz_parse_page_range() to fix SWIG bug where a NULL return # value seems to mess up the returned list - we end up with ret # containing two elements rather than three, e.g. [0, 2]. This # occurs with SWIG-3.0; maybe fixed in SWIG-4? # ll_fz_parse_page_range_orig = ll_fz_parse_page_range def ll_fz_parse_page_range(s, n): ret = ll_fz_parse_page_range_orig(s, n) if len(ret) == 2: return None, 0, 0 else: return ret[0], ret[1], ret[2] fz_parse_page_range = ll_fz_parse_page_range # Provide native python implementation of format_output_path() (-> # fz_format_output_path). # def ll_fz_format_output_path( format, page): m = re.search( '(%[0-9]*d)', format) if m: ret = format[ :m.start(1)] + str(page) + format[ m.end(1):] else: dot = format.rfind( '.') if dot < 0: dot = len( format) ret = format[:dot] + str(page) + format[dot:] return ret fz_format_output_path = ll_fz_format_output_path class IteratorWrap: """ This is a Python iterator for containers that have C++-style begin() and end() methods that return iterators. Iterators must have the following methods: __increment__(): move to next item in the container. __ref__(): return reference to item in the container. Must also be able to compare two iterators for equality. """ def __init__( self, container): self.container = container self.pos = None self.end = container.end() def __iter__( self): return self def __next__( self): # for python2. if self.pos is None: self.pos = self.container.begin() else: self.pos.__increment__() if self.pos == self.end: raise StopIteration() return self.pos.__ref__() def next( self): # for python3. return self.__next__() # The auto-generated Python class method # {rename.class_('fz_buffer')}.{rename.method('fz_buffer', 'fz_buffer_extract')}() returns (size, data). # # But these raw values aren't particularly useful to # Python code so we change the method to return a Python # bytes instance instead, using the special C function # buffer_extract_bytes() defined above. # # The raw values for a buffer are available via # fz_buffer_storage(). def ll_fz_buffer_extract(buffer): """ Returns buffer data as a Python bytes instance, leaving the buffer empty. """ assert isinstance( buffer, fz_buffer) return ll_fz_buffer_to_bytes_internal(buffer, clear=1) def fz_buffer_extract(buffer): """ Returns buffer data as a Python bytes instance, leaving the buffer empty. """ assert isinstance( buffer, FzBuffer) return ll_fz_buffer_extract(buffer.m_internal) {set_class_method('fz_buffer', 'fz_buffer_extract')} def ll_fz_buffer_extract_copy( buffer): """ Returns buffer data as a Python bytes instance, leaving the buffer unchanged. """ assert isinstance( buffer, fz_buffer) return ll_fz_buffer_to_bytes_internal(buffer, clear=0) def fz_buffer_extract_copy( buffer): """ Returns buffer data as a Python bytes instance, leaving the buffer unchanged. """ assert isinstance( buffer, FzBuffer) return ll_fz_buffer_extract_copy(buffer.m_internal) {set_class_method('fz_buffer', 'fz_buffer_extract_copy')} # [ll_fz_buffer_storage_memoryview() is implemented in C.] def fz_buffer_storage_memoryview( buffer, writable=False): """ Returns a read-only or writable Python `memoryview` onto `fz_buffer` data. This relies on `buffer` existing and not changing size while the `memoryview` is used. """ assert isinstance( buffer, FzBuffer) return ll_fz_buffer_storage_memoryview( buffer.m_internal, writable) {set_class_method('fz_buffer', 'fz_buffer_storage_memoryview')} # Overwrite wrappers for fz_new_buffer_from_copied_data() to # take Python buffer. # ll_fz_new_buffer_from_copied_data_orig = ll_fz_new_buffer_from_copied_data def ll_fz_new_buffer_from_copied_data(data): """ Returns fz_buffer containing copy of `data`, which should be a `bytes` or similar Python buffer instance. """ buffer_ = ll_fz_new_buffer_from_copied_data_orig(python_buffer_data(data), len(data)) return buffer_ def fz_new_buffer_from_copied_data(data): """ Returns FzBuffer containing copy of `data`, which should be a `bytes` or similar Python buffer instance. """ return FzBuffer( ll_fz_new_buffer_from_copied_data( data)) {set_class_method('fz_buffer', 'fz_new_buffer_from_copied_data')} def ll_pdf_dict_getl(obj, *tail): """ Python implementation of ll_pdf_dict_getl(), because SWIG doesn't handle variadic args. Each item in `tail` should be `mupdf.pdf_obj`. """ for key in tail: if not obj: break obj = ll_pdf_dict_get(obj, key) assert isinstance(obj, pdf_obj) return obj def pdf_dict_getl(obj, *tail): """ Python implementation of pdf_dict_getl(), because SWIG doesn't handle variadic args. Each item in `tail` should be a `mupdf.PdfObj`. """ for key in tail: if not obj.m_internal: break obj = pdf_dict_get(obj, key) assert isinstance(obj, PdfObj) return obj {set_class_method('pdf_obj', 'pdf_dict_getl')} def ll_pdf_dict_putl(obj, val, *tail): """ Python implementation of ll_pdf_dict_putl() because SWIG doesn't handle variadic args. Each item in `tail` should be a SWIG wrapper for a `pdf_obj`. """ if ll_pdf_is_indirect( obj): obj = ll_pdf_resolve_indirect_chain( obj) if not pdf_is_dict( obj): raise Exception(f'not a dict: {{obj}}') if not tail: return doc = ll_pdf_get_bound_document( obj) for i, key in enumerate( tail[:-1]): assert isinstance( key, PdfObj), f'Item {{i}} in `tail` should be a pdf_obj but is a {{type(key)}}.' next_obj = ll_pdf_dict_get( obj, key) if not next_obj: # We have to create entries next_obj = ll_pdf_new_dict( doc, 1) ll_pdf_dict_put( obj, key, next_obj) obj = next_obj key = tail[-1] ll_pdf_dict_put( obj, key, val) def pdf_dict_putl(obj, val, *tail): """ Python implementation of pdf_dict_putl(fz_context *ctx, pdf_obj *obj, pdf_obj *val, ...) because SWIG doesn't handle variadic args. Each item in `tail` should be a SWIG wrapper for a `PdfObj`. """ if pdf_is_indirect( obj): obj = pdf_resolve_indirect_chain( obj) if not pdf_is_dict( obj): raise Exception(f'not a dict: {{obj}}') if not tail: return doc = pdf_get_bound_document( obj) for i, key in enumerate( tail[:-1]): assert isinstance( key, PdfObj), f'item {{i}} in `tail` should be a PdfObj but is a {{type(key)}}.' next_obj = pdf_dict_get( obj, key) if not next_obj.m_internal: # We have to create entries next_obj = pdf_new_dict( doc, 1) pdf_dict_put( obj, key, next_obj) obj = next_obj key = tail[-1] pdf_dict_put( obj, key, val) {set_class_method('pdf_obj', 'pdf_dict_putl')} def pdf_dict_putl_drop(obj, *tail): raise Exception('mupdf.pdf_dict_putl_drop() is unsupported and unnecessary in Python because reference counting is automatic. Instead use mupdf.pdf_dict_putl().') {set_class_method('pdf_obj', 'pdf_dict_putl_drop')} def ll_pdf_set_annot_color(annot, color): """ Low-level Python implementation of pdf_set_annot_color() using ll_pdf_set_annot_color2(). """ if isinstance(color, float): ll_pdf_set_annot_color2(annot, 1, color, 0, 0, 0) elif len(color) == 1: ll_pdf_set_annot_color2(annot, 1, color[0], 0, 0, 0) elif len(color) == 2: ll_pdf_set_annot_color2(annot, 2, color[0], color[1], 0, 0) elif len(color) == 3: ll_pdf_set_annot_color2(annot, 3, color[0], color[1], color[2], 0) elif len(color) == 4: ll_pdf_set_annot_color2(annot, 4, color[0], color[1], color[2], color[3]) else: raise Exception( f'Unexpected color should be float or list of 1-4 floats: {{color}}') def pdf_set_annot_color(self, color): return ll_pdf_set_annot_color(self.m_internal, color) {set_class_method('pdf_annot', 'pdf_set_annot_color')} def ll_pdf_set_annot_interior_color(annot, color): """ Low-level Python version of pdf_set_annot_color() using pdf_set_annot_color2(). """ if isinstance(color, float): ll_pdf_set_annot_interior_color2(annot, 1, color, 0, 0, 0) elif len(color) == 1: ll_pdf_set_annot_interior_color2(annot, 1, color[0], 0, 0, 0) elif len(color) == 2: ll_pdf_set_annot_interior_color2(annot, 2, color[0], color[1], 0, 0) elif len(color) == 3: ll_pdf_set_annot_interior_color2(annot, 3, color[0], color[1], color[2], 0) elif len(color) == 4: ll_pdf_set_annot_interior_color2(annot, 4, color[0], color[1], color[2], color[3]) else: raise Exception( f'Unexpected color should be float or list of 1-4 floats: {{color}}') def pdf_set_annot_interior_color(self, color): """ Python version of pdf_set_annot_color() using pdf_set_annot_color2(). """ return ll_pdf_set_annot_interior_color(self.m_internal, color) {set_class_method('pdf_annot', 'pdf_set_annot_interior_color')} def ll_fz_fill_text( dev, text, ctm, colorspace, color, alpha, color_params): """ Low-level Python version of fz_fill_text() taking list/tuple for `color`. """ color = tuple(color) + (0,) * (4-len(color)) assert len(color) == 4, f'color not len 4: len={{len(color)}}: {{color}}' return ll_fz_fill_text2(dev, text, ctm, colorspace, *color, alpha, color_params) def fz_fill_text(dev, text, ctm, colorspace, color, alpha, color_params): """ Python version of fz_fill_text() taking list/tuple for `color`. """ return ll_fz_fill_text( dev.m_internal, text.m_internal, ctm.internal(), colorspace.m_internal, color, alpha, color_params.internal(), ) {set_class_method('fz_device', 'fz_fill_text')} # Override mupdf_convert_color() to return (rgb0, rgb1, rgb2, rgb3). def ll_fz_convert_color( ss, sv, ds, is_, params): """ Low-level Python version of fz_convert_color(). `sv` should be a float or list of 1-4 floats or a SWIG representation of a float*. Returns (dv0, dv1, dv2, dv3). """ dv = fz_convert_color2_v() if isinstance( sv, float): ll_fz_convert_color2( ss, sv, 0.0, 0.0, 0.0, ds, dv, is_, params) elif isinstance( sv, (tuple, list)): sv2 = tuple(sv) + (0,) * (4-len(sv)) ll_fz_convert_color2( ss, *sv2, ds, dv, is_, params) else: # Assume `sv` is SWIG representation of a `float*`. ll_fz_convert_color2( ss, sv, ds, dv, is_, params) return dv.v0, dv.v1, dv.v2, dv.v3 def fz_convert_color( ss, sv, ds, is_, params): """ Python version of fz_convert_color(). `sv` should be a float or list of 1-4 floats or a SWIG representation of a float*. Returns (dv0, dv1, dv2, dv3). """ return ll_fz_convert_color( ss.m_internal, sv, ds.m_internal, is_.m_internal, params.internal()) {set_class_method('fz_colorspace', 'fz_convert_color')} # Override fz_set_warning_callback() and # fz_set_error_callback() to use Python classes derived from # our SWIG Director class DiagnosticCallback (defined in C), so # that fnptrs can call Python code. # # We store DiagnosticCallbackPython instances in these # globals to ensure they continue to exist after # set_diagnostic_callback() returns. # set_warning_callback_s = None set_error_callback_s = None # Override set_error_callback(). class DiagnosticCallbackPython( DiagnosticCallback): """ Overrides Director class DiagnosticCallback's virtual `_print()` method in Python. """ def __init__( self, description, printfn): super().__init__( description) self.printfn = printfn if g_mupdf_trace_director: log( f'DiagnosticCallbackPython[{{self.m_description}}].__init__() self={{self!r}} printfn={{printfn!r}}') def __del__( self): if g_mupdf_trace_director: log( f'DiagnosticCallbackPython[{{self.m_description}}].__del__() destructor called.') def _print( self, message): if g_mupdf_trace_director: log( f'DiagnosticCallbackPython[{{self.m_description}}]._print(): Calling self.printfn={{self.printfn!r}} with message={{message!r}}') try: self.printfn( message) except Exception as e: # This shouldn't happen, so always output a diagnostic. log( f'DiagnosticCallbackPython[{{self.m_description}}]._print(): Warning: exception from self.printfn={{self.printfn!r}}: e={{e!r}}') # Calling `raise` here serves to test # `DiagnosticCallback()`'s swallowing of what will # be a C++ exception. But we could swallow the # exception here instead. raise def set_diagnostic_callback( description, printfn): if g_mupdf_trace_director: log( f'set_diagnostic_callback() description={{description!r}} printfn={{printfn!r}}') if printfn: ret = DiagnosticCallbackPython( description, printfn) return ret else: if g_mupdf_trace_director: log( f'Calling ll_fz_set_{{description}}_callback() with (None, None)') if description == 'error': ll_fz_set_error_callback( None, None) elif description == 'warning': ll_fz_set_warning_callback( None, None) else: assert 0, f'Unrecognised description={{description!r}}' return None def fz_set_error_callback( printfn): global set_error_callback_s set_error_callback_s = set_diagnostic_callback( 'error', printfn) def fz_set_warning_callback( printfn): global set_warning_callback_s set_warning_callback_s = set_diagnostic_callback( 'warning', printfn) # Direct access to fz_pixmap samples. def ll_fz_pixmap_samples_memoryview( pixmap): """ Returns a writable Python `memoryview` for a `fz_pixmap`. """ assert isinstance( pixmap, fz_pixmap) ret = python_memoryview_from_memory( ll_fz_pixmap_samples( pixmap), ll_fz_pixmap_stride( pixmap) * ll_fz_pixmap_height( pixmap), 1, # writable ) return ret def fz_pixmap_samples_memoryview( pixmap): """ Returns a writable Python `memoryview` for a `FzPixmap`. """ return ll_fz_pixmap_samples_memoryview( pixmap.m_internal) {set_class_method('fz_pixmap', 'fz_pixmap_samples_memoryview')} # Avoid potential unsafe use of variadic args by forcing a # single arg and escaping all '%' characters. (Passing ('%s', # text) does not work - results in "(null)" being output.) # ll_fz_warn_original = ll_fz_warn def ll_fz_warn( text): assert isinstance( text, str), f'text={{text!r}} str={{str!r}}' text = text.replace( '%', '%%') return ll_fz_warn_original( text) fz_warn = ll_fz_warn # Force use of pdf_load_field_name2() instead of # pdf_load_field_name() because the latter returns a char* # buffer that must be freed by the caller. ll_pdf_load_field_name = ll_pdf_load_field_name2 pdf_load_field_name = pdf_load_field_name2 {set_class_method('pdf_obj', 'pdf_load_field_name')} # It's important that when we create class derived # from StoryPositionsCallback, we ensure that # StoryPositionsCallback's constructor is called. Otherwise # the new instance doesn't seem to be an instance of # StoryPositionsCallback. # class StoryPositionsCallback_python( StoryPositionsCallback): def __init__( self, python_callback): super().__init__() self.python_callback = python_callback def call( self, position): self.python_callback( position) ll_fz_story_positions_orig = ll_fz_story_positions def ll_fz_story_positions( story, python_callback): """ Custom replacement for `ll_fz_story_positions()` that takes a Python callable `python_callback`. """ #log( f'll_fz_story_positions() type(story)={{type(story)!r}} type(python_callback)={{type(python_callback)!r}}') python_callback_instance = StoryPositionsCallback_python( python_callback) ll_fz_story_positions_director( story, python_callback_instance) def fz_story_positions( story, python_callback): #log( f'fz_story_positions() type(story)={{type(story)!r}} type(python_callback)={{type(python_callback)!r}}') assert isinstance( story, FzStory) assert callable( python_callback) def python_callback2( position): position2 = FzStoryElementPosition( position) python_callback( position2) ll_fz_story_positions( story.m_internal, python_callback2) {set_class_method('fz_story', 'fz_story_positions')} # Monkey-patch `FzDocumentWriter.__init__()` to set `self._out` # to any `FzOutput2` arg. This ensures that the Python part of # the derived `FzOutput2` instance is kept alive for use by the # `FzDocumentWriter`, otherwise Python can delete it, then get # a SEGV if C++ tries to call the derived Python methods. # # [We don't patch equivalent class-aware functions such # as `fz_new_pdf_writer_with_output()` because they are # not available to C++/Python, because FzDocumentWriter is # non-copyable.] # FzDocumentWriter__init__0 = FzDocumentWriter.__init__ def FzDocumentWriter__init__1(self, *args): out = None for arg in args: if isinstance( arg, FzOutput2): assert not out, "More than one FzOutput2 passed to FzDocumentWriter.__init__()" out = arg if out is not None: self._out = out return FzDocumentWriter__init__0(self, *args) FzDocumentWriter.__init__ = FzDocumentWriter__init__1 # Create class derived from # fz_install_load_system_font_funcs_args class wrapper with # overrides of the virtual functions to allow calling of Python # callbacks. # class fz_install_load_system_font_funcs_args3({rename.class_('fz_install_load_system_font_funcs_args')}2): """ Class derived from Swig Director class fz_install_load_system_font_funcs_args2, to allow implementation of fz_install_load_system_font_funcs with Python callbacks. """ def __init__(self, f=None, f_cjk=None, f_fallback=None): super().__init__() self.f3 = f self.f_cjk3 = f_cjk self.f_fallback3 = f_fallback self.use_virtual_f(True if f else False) self.use_virtual_f_cjk(True if f_cjk else False) self.use_virtual_f_fallback(True if f_fallback else False) def ret_font(self, font): if font is None: return None elif isinstance(font, {rename.class_('fz_font')}): return ll_fz_keep_font(font.m_internal) elif isinstance(font, fz_font): return font else: assert 0, f'Expected FzFont or fz_font, but fz_install_load_system_font_funcs() callback returned {{type(font)=}}' def f(self, ctx, name, bold, italic, needs_exact_metrics): font = self.f3(name, bold, italic, needs_exact_metrics) return self.ret_font(font) def f_cjk(self, ctx, name, ordering, serif): font = self.f_cjk3(name, ordering, serif) return self.ret_font(font) def f_fallback(self, ctx, script, language, serif, bold, italic): font = self.f_fallback3(script, language, serif, bold, italic) return self.ret_font(font) # We store the most recently created # fz_install_load_system_font_funcs_args in this global so that # it is not cleaned up by Python. g_fz_install_load_system_font_funcs_args = None def fz_install_load_system_font_funcs(f=None, f_cjk=None, f_fallback=None): """ Python override for MuPDF fz_install_load_system_font_funcs() using Swig Director support. Python callbacks are not passed a `ctx` arg, and can return None, a mupdf.fz_font or a mupdf.FzFont. """ global g_fz_install_load_system_font_funcs_args g_fz_install_load_system_font_funcs_args = fz_install_load_system_font_funcs_args3( f, f_cjk, f_fallback, ) fz_install_load_system_font_funcs2(g_fz_install_load_system_font_funcs_args) Py_LIMITED_API = {repr(build_dirs.Py_LIMITED_API) if build_dirs.Py_LIMITED_API else 'None'} ''') # Add __iter__() methods for all classes with begin() and end() methods. # for classname in generated.container_classnames: text += f'{classname}.__iter__ = lambda self: IteratorWrap( self)\n' # For all wrapper classes with a to_string() method, add a __str__() # method to the underlying struct's Python class, which calls # to_string_<structname>(). # # E.g. this allows Python code to print a mupdf.fz_rect instance. # # [We could instead call our generated to_string() and rely on overloading, # but this will end up switching on the type in the SWIG code.] # for struct_name in generated.to_string_structnames: text += f'{struct_name}.__str__ = lambda s: to_string_{struct_name}(s)\n' text += f'{struct_name}.__repr__ = lambda s: to_string_{struct_name}(s)\n' # For all wrapper classes with a to_string() method, add a __str__() method # to the Python wrapper class, which calls the class's to_string() method. # # E.g. this allows Python code to print a mupdf.Rect instance. # for struct_name in generated.to_string_structnames: text += f'{rename.class_(struct_name)}.__str__ = lambda self: self.to_string()\n' text += f'{rename.class_(struct_name)}.__repr__ = lambda self: self.to_string()\n' text += '%}\n' if 1: # lgtm [py/constant-conditional-expression] # This is a horrible hack to avoid swig failing because # include/mupdf/pdf/object.h defines an enum which contains a #include. # # Would like to pre-process files in advance so that swig doesn't see # the #include, but this breaks swig in a different way - swig cannot # cope with some code in system headers. # # So instead we copy include/mupdf/pdf/object.h into # {build_dirs.dir_mupdf}/platform/python/include/mupdf/pdf/object.h, # manually expanding the #include using a Python .replace() call. Then # we specify {build_dirs.dir_mupdf}/platform/python/include as the # first include path so that our modified mupdf/pdf/object.h will get # included in preference to the original. # os.makedirs(f'{build_dirs.dir_mupdf}/platform/python/include/mupdf/pdf', exist_ok=True) with open( f'{build_dirs.dir_mupdf}/include/mupdf/pdf/object.h') as f: o = f.read() with open( f'{build_dirs.dir_mupdf}/include/mupdf/pdf/name-table.h') as f: name_table_h = f.read() oo = o.replace( '#include "mupdf/pdf/name-table.h"\n', name_table_h) assert oo != o jlib.fs_update( oo, f'{build_dirs.dir_mupdf}/platform/python/include/mupdf/pdf/object.h') swig_i = build_dirs.mupdfcpp_swig_i(language) swig_cpp = build_dirs.mupdfcpp_swig_cpp(language) include1 = f'{build_dirs.dir_mupdf}/include/' include2 = f'{build_dirs.dir_mupdf}/platform/c++/include' swig_py = f'{build_dirs.dir_so}/mupdf.py' swig2_i = f'{build_dirs.dir_mupdf}/platform/{language}/mupdfcpp2_swig.i' swig2_cpp = f'{build_dirs.dir_mupdf}/platform/{language}/mupdfcpp2_swig.cpp' swig2_py = f'{build_dirs.dir_so}/mupdf2.py' os.makedirs( f'{build_dirs.dir_mupdf}/platform/{language}', exist_ok=True) os.makedirs( f'{build_dirs.dir_so}', exist_ok=True) util.update_file_regress( text, swig_i, check_regress) jlib.fs_update( '', swig2_i) # Disable some unhelpful SWIG warnings. Must not use -Wall as it overrides # all warning disables. disable_swig_warnings = [ 201, # Warning 201: Unable to find 'stddef.h' 314, # Warning 314: 'print' is a python keyword, renaming to '_print' 302, # Warning 302: Identifier 'pdf_annot_type' redefined (ignored), 312, # Warning 312: Nested union not currently supported (ignored). 321, # Warning 321: 'max' conflicts with a built-in name in python 322, # Warning 322: Redundant redeclaration of 'pdf_annot', 362, # Warning 362: operator= ignored 451, # Warning 451: Setting a const char * variable may leak memory. 503, # Warning 503: Can't wrap 'operator <<' unless renamed to a valid identifier. 512, # Warning 512: Overloaded method mupdf::DrawOptions::internal() const ignored, using non-const method mupdf::DrawOptions::internal() instead. 509, # Warning 509: Overloaded method mupdf::FzAaContext::FzAaContext(::fz_aa_context const) effectively ignored, 560, # Warning 560: Unknown Doxygen command: d. ] disable_swig_warnings = [ '-' + str( x) for x in disable_swig_warnings] disable_swig_warnings = '-w' + ','.join( disable_swig_warnings) # Preserve any existing file `swig_cpp`, so that we can restore the # mtime if SWIG produces an unchanged file. This then avoids unnecessary # recompilation. # # 2022-11-16: Disabled this, because it can result in continuous # unnecessary rebuilds, e.g. if .cpp is older than a mupdf header. # swig_cpp_old = None if 0 and os.path.exists( swig_cpp): swig_cpp_old = f'{swig_cpp}-old' jlib.fs_copy( swig_cpp, swig_cpp_old) if language == 'python': # Maybe use '^' on windows as equivalent to unix '\\' for multiline # ending? def make_command( module, cpp, swig_i): cpp = os.path.relpath( cpp) swig_i = os.path.relpath( swig_i) # We need to predefine MUPDF_FITZ_HEAP_H to disable parsing of # include/mupdf/fitz/heap.h. Otherwise swig's preprocessor seems to # ignore #undef's in include/mupdf/fitz/heap-imp.h then complains # about redefinition of macros in include/mupdf/fitz/heap.h. command = f''' "{swig_command}" {"-D_WIN32" if state_.windows else ""} -c++ {"-doxygen" if swig_major >= 4 else ""} -python -Wextra {disable_swig_warnings} -module {module} -outdir {os.path.relpath(build_dirs.dir_mupdf)}/platform/python -o {cpp} -includeall {os.environ.get('XCXXFLAGS', '')} -I{os.path.relpath(build_dirs.dir_mupdf)}/platform/python/include -I{os.path.relpath(include1)} -I{os.path.relpath(include2)} -ignoremissing -DMUPDF_FITZ_HEAP_H {swig_i} ''' return command def modify_py( path_in, path_out): with open( path_in) as f: text = f.read() # Change all our PDF_ENUM_NAME_* enums so that they are actually # PdfObj instances so that they can be used like any other PdfObj. # #jlib.log('{len(generated.c_enums)=}') for enum_type, enum_names in generated.c_enums.items(): for enum_name in enum_names: if enum_name.startswith( 'PDF_ENUM_NAME_'): text += f'{enum_name} = {rename.class_("pdf_obj")}( obj_enum_to_obj( {enum_name}))\n' # 2024-09-28: important to not include PDF_LIMIT here, because # pdf_drop_obj() treats all pdf_obj*'s as real pointers if they are # >= PDF_LIMIT. for name in ('NULL', 'TRUE', 'FALSE'): text += f'PDF_{name} = {rename.class_("pdf_obj")}( obj_enum_to_obj( PDF_ENUM_{name}))\n' jlib.fs_update(text, path_out) jlib.fs_update( '', swig2_cpp) jlib.fs_remove( swig2_py) # Make main mupdf .so. command = make_command( 'mupdf', swig_cpp, swig_i) swig_py_ = f'{build_dirs.dir_mupdf}/platform/python/mupdf.py' rebuilt = jlib.build( (swig_i, include1, include2), (swig_cpp, swig_py_), command, force_rebuild, ) jlib.log(f'swig => {rebuilt=}.') updated = modify_py( swig_py_, swig_py) jlib.log(f'modify_py() => {updated=}.') elif language == 'csharp': outdir = os.path.relpath(f'{build_dirs.dir_mupdf}/platform/csharp') os.makedirs(outdir, exist_ok=True) # Looks like swig comes up with 'mupdfcpp_swig_wrap.cxx' leafname. # # We include platform/python/include in order to pick up the modified # include/mupdf/pdf/object.h that we generate elsewhere. dllimport = 'mupdfcsharp.so' if state_.windows: # Would like to specify relative path to .dll with: # dllimport = os.path.relpath( f'{build_dirs.dir_so}/mupdfcsharp.dll') # but Windows/.NET doesn't seem to support this, despite # https://stackoverflow.com/questions/31807289 "how can i add a # swig generated c dll reference to a c sharp project". # dllimport = 'mupdfcsharp.dll' # See https://www.swig.org/Doc4.2/CSharp.html `23.3.1 Primitive types` # for description of SWIGWORDSIZE64. If we were to build on 32-bit Linux # we would have to remove the `-DSWIGWORDSIZE64` flag. command = (f''' "{swig_command}" {"-D_WIN32" if state_.windows else ""} {"-DSWIGWORDSIZE64" if state_.linux else ""} -c++ -csharp -Wextra {disable_swig_warnings} -module mupdf -namespace mupdf -dllimport {dllimport} -outdir {outdir} -outfile mupdf.cs -o {os.path.relpath(swig_cpp)} -includeall -I{os.path.relpath(build_dirs.dir_mupdf)}/platform/python/include -I{os.path.relpath(include1)} -I{os.path.relpath(include2)} -ignoremissing -DMUPDF_FITZ_HEAP_H {os.path.relpath(swig_i)} ''') rebuilt = jlib.build( (swig_i, include1, include2), (f'{outdir}/mupdf.cs', os.path.relpath(swig_cpp)), command, force_rebuild, ) # fixme: use <rebuilt> line with language=='python' to avoid multiple # modifications to unchanged mupdf.cs? # # For classes that have our to_string() method, override C#'s # ToString() to call to_string(). with open(f'{outdir}/mupdf.cs') as f: cs = f.read() cs2 = re.sub( '(( *)public string to_string[(][)])', '\\2public override string ToString() { return to_string(); }\n\\1', cs, ) jlib.log1('{len(cs)=}') jlib.log1('{len(cs2)=}') assert cs2 != cs, f'Failed to add toString() methods.' jlib.log1('{len(generated.swig_csharp)=}') assert len(generated.swig_csharp) cs2 += generated.swig_csharp jlib.log1( 'Updating cs2 => {build_dirs.dir_so}/mupdf.cs') jlib.fs_update(cs2, f'{build_dirs.dir_so}/mupdf.cs') #jlib.fs_copy(f'{outdir}/mupdf.cs', f'{build_dirs.dir_so}/mupdf.cs') jlib.log1('{rebuilt=}') else: assert 0 # Disabled; see above for explanation. if 0 and swig_cpp_old: with open( swig_cpp_old) as f: swig_cpp_contents_old = f.read() with open(swig_cpp) as f: swig_cpp_contents_new = f.read() if swig_cpp_contents_new == swig_cpp_contents_old: # File <swig_cpp> unchanged, so restore the mtime to avoid # unnecessary recompilation. jlib.log( 'File contents unchanged, copying {swig_cpp_old=} => {swig_cpp=}') jlib.fs_rename( swig_cpp_old, swig_cpp) def test_swig(): ''' For testing different swig .i constructs. ''' test_i = textwrap.dedent(''' %include argcargv.i %apply (int ARGC, char **ARGV) { (int retainlen, const char **retainlist) } %apply (int ARGC, char **ARGV) { (const char **retainlist, int retainlen) } %apply (int ARGC, char **ARGV) { (const char *retainlist[], int retainlen) } %clear double a, int ARGC, char **ARGV; %clear double a, int argc, char *argv[]; %clear int ARGC, char **ARGV; %clear (double a, int ARGC, char **ARGV); %clear (double a, int argc, char *argv[]); %clear (int ARGC, char **ARGV); %clear int retainlen, const char **retainlist; int bar( int argc, char* argv[]); int foo( double a, int argc, char* argv[]); int qwe( double a, int argc, const char** argv); void ppdf_clean_file( char *infile, char *outfile, char *password, pdf_write_options *opts, int retainlen, const char **retainlist); void ppdf_clean_file2(char *infile, char *outfile, char *password, pdf_write_options *opts, const char **retainlist, int retainlen); void ppdf_clean_file3(char *infile, char *outfile, char *password, pdf_write_options *opts, const char *retainlist[], int retainlen); ''') jlib.fs_update( test_i, 'test.i') jlib.system( textwrap.dedent( ''' swig -Wall -c++ -python -module test -outdir . -o test.cpp test.i ''').replace( '\n', ' \\\n') ) def test_swig_csharp(): ''' Checks behaviour with and without our custom string marshalling code from _csharp_unicode_prefix(). ''' test_swig_csharp_internal(fix=0) test_swig_csharp_internal(fix=1) def test_swig_csharp_internal(fix): ''' Test utf8 string handling, with/without use of _csharp_unicode_prefix(). ''' # We create C++/C# source directly from this function, and explicitly run # C++ and .NET/Mono build commands. # build_dir = f'test_swig_{fix}' os.makedirs( build_dir, exist_ok=True) print('') print(f'### test_swig_internal(): {fix=}', flush=1) # Create SWIG input file `test.i`. # test_i = '%module test\n' if fix: test_i += _csharp_unicode_prefix() test_i += textwrap.dedent(f''' %include "std_string.i" // Returns escaped representation of `text`. const char* foo1(const char* text); // Returns escaped representation of `text`. std::string foo2(const std::string& text); // Returns 4-byte string `0xf0 0x90 0x90 0xb7`, which decodes as // utf8 to a 4-byte utf16 character. const char* bar(); // Returns 4-byte string `0xf0 0x90 0x90 0xb7`, which decodes as // utf8 to a 4-byte utf16 character. std::string bar2(); %{{ // Returns string containing escaped description of `text`. std::string foo2(const std::string& text) {{ std::string ret; for (int i=0; i<text.size(); ++i) {{ char buffer[8]; snprintf(buffer, sizeof(buffer), " \\\\x%02x", (unsigned char) text[i]); ret += buffer; }} return ret; }} // Returns pointer to static buffer containing escaped // description of `text`. const char* foo1(const char* text) {{ std::string text2 = text; static std::string ret; ret = foo2(text2); return ret.c_str(); }} // Returns pointer to static buffer containing a utf8 string. const char* bar() {{ static char ret[] = {{ (char) 0xf0, (char) 0x90, (char) 0x90, (char) 0xb7, 0, }}; return ret; }} // Returns a std::string containing a utf8 string. std::string bar2() {{ const char* ret = bar(); return std::string(ret); }} %}} ''') with open(f'{build_dir}/test.i', 'w') as f: f.write(test_i) # Run swig on `test.i` to generate `test.cs` and `test.cpp`. # jlib.system( f''' cd {build_dir} && swig {'-DSWIG_CSHARP_NO_STRING_HELPER=1 -DSWIG_CSHARP_NO_EXCEPTION_HELPER=1' if 0 and fix else ''} -D_WIN32 -c++ -csharp -Wextra -Wall -dllimport test.dll -outdir . -outfile test.cs -o test.cpp test.i ''') # Compile/link test.cpp to create test.dll. # if state.state_.windows: import wdev vs = wdev.WindowsVS() jlib.system( f''' cd {build_dir} && "{vs.vcvars}"&&"{vs.cl}" /nologo # /c # Compiles without linking. /EHsc # Enable "Standard C++ exception handling". /MD /Tptest.cpp # /Tp specifies C++ source file. /Fotest.cpp.obj # Output file. /permissive- # Set standard-conformance mode. /FC # Display full path of source code files passed to cl.exe in diagnostic text. /W3 # Sets which warning level to output. /W3 is IDE default. /diagnostics:caret # Controls the format of diagnostic messages. ''') jlib.system( f''' cd {build_dir} && "{vs.vcvars}"&&"{vs.link}" /nologo # /DLL /IMPLIB:test.lib # Overrides the default import library name. /OUT:test.dll # Specifies the output file name. /nologo test.cpp.obj ''') else: jlib.system( f''' cd {build_dir} && c++ -fPIC --shared -o test.dll test.cpp ''') # Create C# test programme `testfoo.cs`. # cs = textwrap.dedent(f''' public class HelloWorld {{ public static void Main(string[] args) {{ bool expect_fix = ({fix if state.state_.windows else 1} != 0); // Utf8 for our string with 4-byte utf16 character. // byte[] text_utf8 = {{ 0xf0, 0x90, 0x90, 0xb7, }}; string text = System.Text.Encoding.UTF8.GetString(text_utf8); // Escaped representation of text_utf8, as returned by // calls of test.foo1() and test.foo2() below. // string text_utf8_escaped = " \\\\xf0 \\\\x90 \\\\x90 \\\\xb7"; string incorrect_utf8_escaped = " \\\\x3f \\\\x3f"; // test.foo1()/test.foo2() return a `const // char*`/`std::string` containing an escaped // representation of the string that they were given. If // things are working correctly, this will be an escaped // representation of `text_utf8`. // string foo1 = test.foo1(text); System.Console.WriteLine("foo1: " + foo1); string foo_expected_escaped = (expect_fix) ? text_utf8_escaped : incorrect_utf8_escaped; if (foo1 != foo_expected_escaped) {{ throw new System.Exception( "foo1 incorrect: '" + foo1 + "'" + " - foo_expected_escaped: '" + foo_expected_escaped + "'" ); }} string foo2 = test.foo2(text); System.Console.WriteLine("foo2: " + foo2); if (foo2 != foo_expected_escaped) {{ throw new System.Exception( "foo2 incorrect: '" + foo2 + "'" + " - foo_expected_escaped: '" + foo_expected_escaped + "'" ); }} // test.bar1() and test.bar2() return a `const // char*`/`std::string` containing the bytes of // `text_utf8`. If things are working correctly we will see // exactly these bytes. // byte[] bar_expected_utf8_incorrect = {{ 0xc3, 0xb0, 0xc2, 0x90, 0xc2, 0x90, 0xc2, 0xb7, }}; byte[] bar_expected_utf8 = (expect_fix) ? text_utf8 : bar_expected_utf8_incorrect; string ret3 = test.bar(); byte[] ret3_utf8 = System.Text.Encoding.UTF8.GetBytes(ret3); print_bytes_as_string("ret3_utf8:", ret3_utf8); if (!equal(ret3_utf8, bar_expected_utf8)) {{ throw new System.Exception("ret3 != bar_expected_utf8"); }} string ret4 = test.bar2(); byte[] ret4_utf8 = System.Text.Encoding.UTF8.GetBytes(ret4); print_bytes_as_string("ret4_utf8:", ret4_utf8); if (!equal(ret4_utf8, bar_expected_utf8)) {{ throw new System.Exception("ret4_utf8 != bar_expected_utf8"); }} }} static bool equal(byte[] a, byte[] b) {{ if (a.Length != b.Length) return false; for (int i=0; i<a.Length; ++i) {{ if (a[i] != b[i]) return false; }} return true; }} static void print_bytes_as_string(string prefix, byte[] a) {{ System.Console.Write(prefix); System.Console.Write("["); foreach (var b in a) {{ System.Console.Write(" {{0:x2}}", b); }} System.Console.WriteLine("]"); }} }} ''') with open(f'{build_dir}/testfoo.cs', 'w') as f: f.write(cs) # Use `csc` to compile `testfoo.cs` and create `testfoo.exe`. # csc, mono, _ = csharp.csharp_settings(None) jlib.system(f'cd {build_dir} && "{csc}" -out:testfoo.exe testfoo.cs test.cs') # Run `testfoo.exe`. # jlib.system(f'cd {build_dir} && {mono} testfoo.exe')