Tagging for release of Python 2.5b1
[python/dscho.git] / PC / getpathp.c
blob2bd3f2653fda79ca4f64400e2d1ca66f34b19638
2 /* Return the initial module search path. */
3 /* Used by DOS, OS/2, Windows 3.1, Windows 95/98, Windows NT. */
5 /* ----------------------------------------------------------------
6 PATH RULES FOR WINDOWS:
7 This describes how sys.path is formed on Windows. It describes the
8 functionality, not the implementation (ie, the order in which these
9 are actually fetched is different)
11 * Python always adds an empty entry at the start, which corresponds
12 to the current directory.
14 * If the PYTHONPATH env. var. exists, its entries are added next.
16 * We look in the registry for "application paths" - that is, sub-keys
17 under the main PythonPath registry key. These are added next (the
18 order of sub-key processing is undefined).
19 HKEY_CURRENT_USER is searched and added first.
20 HKEY_LOCAL_MACHINE is searched and added next.
21 (Note that all known installers only use HKLM, so HKCU is typically
22 empty)
24 * We attempt to locate the "Python Home" - if the PYTHONHOME env var
25 is set, we believe it. Otherwise, we use the path of our host .EXE's
26 to try and locate our "landmark" (lib\\os.py) and deduce our home.
27 - If we DO have a Python Home: The relevant sub-directories (Lib,
28 plat-win, lib-tk, etc) are based on the Python Home
29 - If we DO NOT have a Python Home, the core Python Path is
30 loaded from the registry. This is the main PythonPath key,
31 and both HKLM and HKCU are combined to form the path)
33 * Iff - we can not locate the Python Home, have not had a PYTHONPATH
34 specified, and can't locate any Registry entries (ie, we have _nothing_
35 we can assume is a good path), a default path with relative entries is
36 used (eg. .\Lib;.\plat-win, etc)
39 The end result of all this is:
40 * When running python.exe, or any other .exe in the main Python directory
41 (either an installed version, or directly from the PCbuild directory),
42 the core path is deduced, and the core paths in the registry are
43 ignored. Other "application paths" in the registry are always read.
45 * When Python is hosted in another exe (different directory, embedded via
46 COM, etc), the Python Home will not be deduced, so the core path from
47 the registry is used. Other "application paths" in the registry are
48 always read.
50 * If Python can't find its home and there is no registry (eg, frozen
51 exe, some very strange installation setup) you get a path with
52 some default, but relative, paths.
54 ---------------------------------------------------------------- */
57 #include "Python.h"
58 #include "osdefs.h"
60 #ifdef MS_WINDOWS
61 #include <windows.h>
62 #include <tchar.h>
63 #endif
65 #ifdef HAVE_SYS_TYPES_H
66 #include <sys/types.h>
67 #endif /* HAVE_SYS_TYPES_H */
69 #ifdef HAVE_SYS_STAT_H
70 #include <sys/stat.h>
71 #endif /* HAVE_SYS_STAT_H */
73 #include <string.h>
75 /* Search in some common locations for the associated Python libraries.
77 * Py_GetPath() tries to return a sensible Python module search path.
79 * The approach is an adaptation for Windows of the strategy used in
80 * ../Modules/getpath.c; it uses the Windows Registry as one of its
81 * information sources.
84 #ifndef LANDMARK
85 #define LANDMARK "lib\\os.py"
86 #endif
88 static char prefix[MAXPATHLEN+1];
89 static char progpath[MAXPATHLEN+1];
90 static char dllpath[MAXPATHLEN+1];
91 static char *module_search_path = NULL;
94 static int
95 is_sep(char ch) /* determine if "ch" is a separator character */
97 #ifdef ALTSEP
98 return ch == SEP || ch == ALTSEP;
99 #else
100 return ch == SEP;
101 #endif
104 /* assumes 'dir' null terminated in bounds. Never writes
105 beyond existing terminator.
107 static void
108 reduce(char *dir)
110 size_t i = strlen(dir);
111 while (i > 0 && !is_sep(dir[i]))
112 --i;
113 dir[i] = '\0';
117 static int
118 exists(char *filename)
120 struct stat buf;
121 return stat(filename, &buf) == 0;
124 /* Assumes 'filename' MAXPATHLEN+1 bytes long -
125 may extend 'filename' by one character.
127 static int
128 ismodule(char *filename) /* Is module -- check for .pyc/.pyo too */
130 if (exists(filename))
131 return 1;
133 /* Check for the compiled version of prefix. */
134 if (strlen(filename) < MAXPATHLEN) {
135 strcat(filename, Py_OptimizeFlag ? "o" : "c");
136 if (exists(filename))
137 return 1;
139 return 0;
142 /* Add a path component, by appending stuff to buffer.
143 buffer must have at least MAXPATHLEN + 1 bytes allocated, and contain a
144 NUL-terminated string with no more than MAXPATHLEN characters (not counting
145 the trailing NUL). It's a fatal error if it contains a string longer than
146 that (callers must be careful!). If these requirements are met, it's
147 guaranteed that buffer will still be a NUL-terminated string with no more
148 than MAXPATHLEN characters at exit. If stuff is too long, only as much of
149 stuff as fits will be appended.
151 static void
152 join(char *buffer, char *stuff)
154 size_t n, k;
155 if (is_sep(stuff[0]))
156 n = 0;
157 else {
158 n = strlen(buffer);
159 if (n > 0 && !is_sep(buffer[n-1]) && n < MAXPATHLEN)
160 buffer[n++] = SEP;
162 if (n > MAXPATHLEN)
163 Py_FatalError("buffer overflow in getpathp.c's joinpath()");
164 k = strlen(stuff);
165 if (n + k > MAXPATHLEN)
166 k = MAXPATHLEN - n;
167 strncpy(buffer+n, stuff, k);
168 buffer[n+k] = '\0';
171 /* gotlandmark only called by search_for_prefix, which ensures
172 'prefix' is null terminated in bounds. join() ensures
173 'landmark' can not overflow prefix if too long.
175 static int
176 gotlandmark(char *landmark)
178 int ok;
179 Py_ssize_t n;
181 n = strlen(prefix);
182 join(prefix, landmark);
183 ok = ismodule(prefix);
184 prefix[n] = '\0';
185 return ok;
188 /* assumes argv0_path is MAXPATHLEN+1 bytes long, already \0 term'd.
189 assumption provided by only caller, calculate_path() */
190 static int
191 search_for_prefix(char *argv0_path, char *landmark)
193 /* Search from argv0_path, until landmark is found */
194 strcpy(prefix, argv0_path);
195 do {
196 if (gotlandmark(landmark))
197 return 1;
198 reduce(prefix);
199 } while (prefix[0]);
200 return 0;
203 #ifdef MS_WINDOWS
205 /* a string loaded from the DLL at startup.*/
206 extern const char *PyWin_DLLVersionString;
209 /* Load a PYTHONPATH value from the registry.
210 Load from either HKEY_LOCAL_MACHINE or HKEY_CURRENT_USER.
212 Works in both Unicode and 8bit environments. Only uses the
213 Ex family of functions so it also works with Windows CE.
215 Returns NULL, or a pointer that should be freed.
217 XXX - this code is pretty strange, as it used to also
218 work on Win16, where the buffer sizes werent available
219 in advance. It could be simplied now Win16/Win32s is dead!
222 static char *
223 getpythonregpath(HKEY keyBase, int skipcore)
225 HKEY newKey = 0;
226 DWORD dataSize = 0;
227 DWORD numKeys = 0;
228 LONG rc;
229 char *retval = NULL;
230 TCHAR *dataBuf = NULL;
231 static const TCHAR keyPrefix[] = _T("Software\\Python\\PythonCore\\");
232 static const TCHAR keySuffix[] = _T("\\PythonPath");
233 size_t versionLen;
234 DWORD index;
235 TCHAR *keyBuf = NULL;
236 TCHAR *keyBufPtr;
237 TCHAR **ppPaths = NULL;
239 /* Tried to use sysget("winver") but here is too early :-( */
240 versionLen = _tcslen(PyWin_DLLVersionString);
241 /* Space for all the chars, plus one \0 */
242 keyBuf = keyBufPtr = malloc(sizeof(keyPrefix) +
243 sizeof(TCHAR)*(versionLen-1) +
244 sizeof(keySuffix));
245 if (keyBuf==NULL) goto done;
247 memcpy(keyBufPtr, keyPrefix, sizeof(keyPrefix)-sizeof(TCHAR));
248 keyBufPtr += sizeof(keyPrefix)/sizeof(TCHAR) - 1;
249 memcpy(keyBufPtr, PyWin_DLLVersionString, versionLen * sizeof(TCHAR));
250 keyBufPtr += versionLen;
251 /* NULL comes with this one! */
252 memcpy(keyBufPtr, keySuffix, sizeof(keySuffix));
253 /* Open the root Python key */
254 rc=RegOpenKeyEx(keyBase,
255 keyBuf, /* subkey */
256 0, /* reserved */
257 KEY_READ,
258 &newKey);
259 if (rc!=ERROR_SUCCESS) goto done;
260 /* Find out how big our core buffer is, and how many subkeys we have */
261 rc = RegQueryInfoKey(newKey, NULL, NULL, NULL, &numKeys, NULL, NULL,
262 NULL, NULL, &dataSize, NULL, NULL);
263 if (rc!=ERROR_SUCCESS) goto done;
264 if (skipcore) dataSize = 0; /* Only count core ones if we want them! */
265 /* Allocate a temp array of char buffers, so we only need to loop
266 reading the registry once
268 ppPaths = malloc( sizeof(TCHAR *) * numKeys );
269 if (ppPaths==NULL) goto done;
270 memset(ppPaths, 0, sizeof(TCHAR *) * numKeys);
271 /* Loop over all subkeys, allocating a temp sub-buffer. */
272 for(index=0;index<numKeys;index++) {
273 TCHAR keyBuf[MAX_PATH+1];
274 HKEY subKey = 0;
275 DWORD reqdSize = MAX_PATH+1;
276 /* Get the sub-key name */
277 DWORD rc = RegEnumKeyEx(newKey, index, keyBuf, &reqdSize,
278 NULL, NULL, NULL, NULL );
279 if (rc!=ERROR_SUCCESS) goto done;
280 /* Open the sub-key */
281 rc=RegOpenKeyEx(newKey,
282 keyBuf, /* subkey */
283 0, /* reserved */
284 KEY_READ,
285 &subKey);
286 if (rc!=ERROR_SUCCESS) goto done;
287 /* Find the value of the buffer size, malloc, then read it */
288 RegQueryValueEx(subKey, NULL, 0, NULL, NULL, &reqdSize);
289 if (reqdSize) {
290 ppPaths[index] = malloc(reqdSize);
291 if (ppPaths[index]) {
292 RegQueryValueEx(subKey, NULL, 0, NULL,
293 (LPBYTE)ppPaths[index],
294 &reqdSize);
295 dataSize += reqdSize + 1; /* 1 for the ";" */
298 RegCloseKey(subKey);
300 /* original datasize from RegQueryInfo doesn't include the \0 */
301 dataBuf = malloc((dataSize+1) * sizeof(TCHAR));
302 if (dataBuf) {
303 TCHAR *szCur = dataBuf;
304 DWORD reqdSize = dataSize;
305 /* Copy our collected strings */
306 for (index=0;index<numKeys;index++) {
307 if (index > 0) {
308 *(szCur++) = _T(';');
309 dataSize--;
311 if (ppPaths[index]) {
312 Py_ssize_t len = _tcslen(ppPaths[index]);
313 _tcsncpy(szCur, ppPaths[index], len);
314 szCur += len;
315 assert(dataSize > (DWORD)len);
316 dataSize -= (DWORD)len;
319 if (skipcore)
320 *szCur = '\0';
321 else {
322 /* If we have no values, we dont need a ';' */
323 if (numKeys) {
324 *(szCur++) = _T(';');
325 dataSize--;
327 /* Now append the core path entries -
328 this will include the NULL
330 rc = RegQueryValueEx(newKey, NULL, 0, NULL,
331 (LPBYTE)szCur, &dataSize);
333 /* And set the result - caller must free
334 If MBCS, it is fine as is. If Unicode, allocate new
335 buffer and convert.
337 #ifdef UNICODE
338 retval = (char *)malloc(reqdSize+1);
339 if (retval)
340 WideCharToMultiByte(CP_ACP, 0,
341 dataBuf, -1, /* source */
342 retval, reqdSize+1, /* dest */
343 NULL, NULL);
344 free(dataBuf);
345 #else
346 retval = dataBuf;
347 #endif
349 done:
350 /* Loop freeing my temp buffers */
351 if (ppPaths) {
352 for(index=0;index<numKeys;index++)
353 if (ppPaths[index]) free(ppPaths[index]);
354 free(ppPaths);
356 if (newKey)
357 RegCloseKey(newKey);
358 if (keyBuf)
359 free(keyBuf);
360 return retval;
362 #endif /* MS_WINDOWS */
364 static void
365 get_progpath(void)
367 extern char *Py_GetProgramName(void);
368 char *path = getenv("PATH");
369 char *prog = Py_GetProgramName();
371 #ifdef MS_WINDOWS
372 extern HANDLE PyWin_DLLhModule;
373 #ifdef UNICODE
374 WCHAR wprogpath[MAXPATHLEN+1];
375 /* Windows documents that GetModuleFileName() will "truncate",
376 but makes no mention of the null terminator. Play it safe.
377 PLUS Windows itself defines MAX_PATH as the same, but anyway...
379 wprogpath[MAXPATHLEN]=_T('\0');
380 if (PyWin_DLLhModule &&
381 GetModuleFileName(PyWin_DLLhModule, wprogpath, MAXPATHLEN)) {
382 WideCharToMultiByte(CP_ACP, 0,
383 wprogpath, -1,
384 dllpath, MAXPATHLEN+1,
385 NULL, NULL);
387 wprogpath[MAXPATHLEN]=_T('\0');
388 if (GetModuleFileName(NULL, wprogpath, MAXPATHLEN)) {
389 WideCharToMultiByte(CP_ACP, 0,
390 wprogpath, -1,
391 progpath, MAXPATHLEN+1,
392 NULL, NULL);
393 return;
395 #else
396 /* static init of progpath ensures final char remains \0 */
397 if (PyWin_DLLhModule)
398 if (!GetModuleFileName(PyWin_DLLhModule, dllpath, MAXPATHLEN))
399 dllpath[0] = 0;
400 if (GetModuleFileName(NULL, progpath, MAXPATHLEN))
401 return;
402 #endif
403 #endif
404 if (prog == NULL || *prog == '\0')
405 prog = "python";
407 /* If there is no slash in the argv0 path, then we have to
408 * assume python is on the user's $PATH, since there's no
409 * other way to find a directory to start the search from. If
410 * $PATH isn't exported, you lose.
412 #ifdef ALTSEP
413 if (strchr(prog, SEP) || strchr(prog, ALTSEP))
414 #else
415 if (strchr(prog, SEP))
416 #endif
417 strncpy(progpath, prog, MAXPATHLEN);
418 else if (path) {
419 while (1) {
420 char *delim = strchr(path, DELIM);
422 if (delim) {
423 size_t len = delim - path;
424 /* ensure we can't overwrite buffer */
425 len = min(MAXPATHLEN,len);
426 strncpy(progpath, path, len);
427 *(progpath + len) = '\0';
429 else
430 strncpy(progpath, path, MAXPATHLEN);
432 /* join() is safe for MAXPATHLEN+1 size buffer */
433 join(progpath, prog);
434 if (exists(progpath))
435 break;
437 if (!delim) {
438 progpath[0] = '\0';
439 break;
441 path = delim + 1;
444 else
445 progpath[0] = '\0';
448 static void
449 calculate_path(void)
451 char argv0_path[MAXPATHLEN+1];
452 char *buf;
453 size_t bufsz;
454 char *pythonhome = Py_GetPythonHome();
455 char *envpath = Py_GETENV("PYTHONPATH");
457 #ifdef MS_WINDOWS
458 int skiphome, skipdefault;
459 char *machinepath = NULL;
460 char *userpath = NULL;
461 char zip_path[MAXPATHLEN+1];
462 size_t len;
463 #endif
465 get_progpath();
466 /* progpath guaranteed \0 terminated in MAXPATH+1 bytes. */
467 strcpy(argv0_path, progpath);
468 reduce(argv0_path);
469 if (pythonhome == NULL || *pythonhome == '\0') {
470 if (search_for_prefix(argv0_path, LANDMARK))
471 pythonhome = prefix;
472 else
473 pythonhome = NULL;
475 else
476 strncpy(prefix, pythonhome, MAXPATHLEN);
478 if (envpath && *envpath == '\0')
479 envpath = NULL;
482 #ifdef MS_WINDOWS
483 /* Calculate zip archive path */
484 if (dllpath[0]) /* use name of python DLL */
485 strncpy(zip_path, dllpath, MAXPATHLEN);
486 else /* use name of executable program */
487 strncpy(zip_path, progpath, MAXPATHLEN);
488 zip_path[MAXPATHLEN] = '\0';
489 len = strlen(zip_path);
490 if (len > 4) {
491 zip_path[len-3] = 'z'; /* change ending to "zip" */
492 zip_path[len-2] = 'i';
493 zip_path[len-1] = 'p';
495 else {
496 zip_path[0] = 0;
499 skiphome = pythonhome==NULL ? 0 : 1;
500 machinepath = getpythonregpath(HKEY_LOCAL_MACHINE, skiphome);
501 userpath = getpythonregpath(HKEY_CURRENT_USER, skiphome);
502 /* We only use the default relative PYTHONPATH if we havent
503 anything better to use! */
504 skipdefault = envpath!=NULL || pythonhome!=NULL || \
505 machinepath!=NULL || userpath!=NULL;
506 #endif
508 /* We need to construct a path from the following parts.
509 (1) the PYTHONPATH environment variable, if set;
510 (2) for Win32, the zip archive file path;
511 (3) for Win32, the machinepath and userpath, if set;
512 (4) the PYTHONPATH config macro, with the leading "."
513 of each component replaced with pythonhome, if set;
514 (5) the directory containing the executable (argv0_path).
515 The length calculation calculates #4 first.
516 Extra rules:
517 - If PYTHONHOME is set (in any way) item (3) is ignored.
518 - If registry values are used, (4) and (5) are ignored.
521 /* Calculate size of return buffer */
522 if (pythonhome != NULL) {
523 char *p;
524 bufsz = 1;
525 for (p = PYTHONPATH; *p; p++) {
526 if (*p == DELIM)
527 bufsz++; /* number of DELIM plus one */
529 bufsz *= strlen(pythonhome);
531 else
532 bufsz = 0;
533 bufsz += strlen(PYTHONPATH) + 1;
534 bufsz += strlen(argv0_path) + 1;
535 #ifdef MS_WINDOWS
536 if (userpath)
537 bufsz += strlen(userpath) + 1;
538 if (machinepath)
539 bufsz += strlen(machinepath) + 1;
540 bufsz += strlen(zip_path) + 1;
541 #endif
542 if (envpath != NULL)
543 bufsz += strlen(envpath) + 1;
545 module_search_path = buf = malloc(bufsz);
546 if (buf == NULL) {
547 /* We can't exit, so print a warning and limp along */
548 fprintf(stderr, "Can't malloc dynamic PYTHONPATH.\n");
549 if (envpath) {
550 fprintf(stderr, "Using environment $PYTHONPATH.\n");
551 module_search_path = envpath;
553 else {
554 fprintf(stderr, "Using default static path.\n");
555 module_search_path = PYTHONPATH;
557 #ifdef MS_WINDOWS
558 if (machinepath)
559 free(machinepath);
560 if (userpath)
561 free(userpath);
562 #endif /* MS_WINDOWS */
563 return;
566 if (envpath) {
567 strcpy(buf, envpath);
568 buf = strchr(buf, '\0');
569 *buf++ = DELIM;
571 #ifdef MS_WINDOWS
572 if (zip_path[0]) {
573 strcpy(buf, zip_path);
574 buf = strchr(buf, '\0');
575 *buf++ = DELIM;
577 if (userpath) {
578 strcpy(buf, userpath);
579 buf = strchr(buf, '\0');
580 *buf++ = DELIM;
581 free(userpath);
583 if (machinepath) {
584 strcpy(buf, machinepath);
585 buf = strchr(buf, '\0');
586 *buf++ = DELIM;
587 free(machinepath);
589 if (pythonhome == NULL) {
590 if (!skipdefault) {
591 strcpy(buf, PYTHONPATH);
592 buf = strchr(buf, '\0');
595 #else
596 if (pythonhome == NULL) {
597 strcpy(buf, PYTHONPATH);
598 buf = strchr(buf, '\0');
600 #endif /* MS_WINDOWS */
601 else {
602 char *p = PYTHONPATH;
603 char *q;
604 size_t n;
605 for (;;) {
606 q = strchr(p, DELIM);
607 if (q == NULL)
608 n = strlen(p);
609 else
610 n = q-p;
611 if (p[0] == '.' && is_sep(p[1])) {
612 strcpy(buf, pythonhome);
613 buf = strchr(buf, '\0');
614 p++;
615 n--;
617 strncpy(buf, p, n);
618 buf += n;
619 if (q == NULL)
620 break;
621 *buf++ = DELIM;
622 p = q+1;
625 if (argv0_path) {
626 *buf++ = DELIM;
627 strcpy(buf, argv0_path);
628 buf = strchr(buf, '\0');
630 *buf = '\0';
631 /* Now to pull one last hack/trick. If sys.prefix is
632 empty, then try and find it somewhere on the paths
633 we calculated. We scan backwards, as our general policy
634 is that Python core directories are at the *end* of
635 sys.path. We assume that our "lib" directory is
636 on the path, and that our 'prefix' directory is
637 the parent of that.
639 if (*prefix=='\0') {
640 char lookBuf[MAXPATHLEN+1];
641 char *look = buf - 1; /* 'buf' is at the end of the buffer */
642 while (1) {
643 Py_ssize_t nchars;
644 char *lookEnd = look;
645 /* 'look' will end up one character before the
646 start of the path in question - even if this
647 is one character before the start of the buffer
649 while (*look != DELIM && look >= module_search_path)
650 look--;
651 nchars = lookEnd-look;
652 strncpy(lookBuf, look+1, nchars);
653 lookBuf[nchars] = '\0';
654 /* Up one level to the parent */
655 reduce(lookBuf);
656 if (search_for_prefix(lookBuf, LANDMARK)) {
657 break;
659 /* If we are out of paths to search - give up */
660 if (look < module_search_path)
661 break;
662 look--;
668 /* External interface */
670 char *
671 Py_GetPath(void)
673 if (!module_search_path)
674 calculate_path();
675 return module_search_path;
678 char *
679 Py_GetPrefix(void)
681 if (!module_search_path)
682 calculate_path();
683 return prefix;
686 char *
687 Py_GetExecPrefix(void)
689 return Py_GetPrefix();
692 char *
693 Py_GetProgramFullPath(void)
695 if (!module_search_path)
696 calculate_path();
697 return progpath;