Actual source code: errtrace.c
1: #define PETSC_DESIRE_FEATURE_TEST_MACROS /* for fileno() */
2: #include <petscsys.h>
3: #include <petsc/private/petscimpl.h>
4: #include <petscconfiginfo.h>
5: #if defined(PETSC_HAVE_UNISTD_H)
6: #include <unistd.h>
7: #endif
8: #include "err.h"
9: #include <petsc/private/logimpl.h>
11: #if defined(PETSC_HAVE_CUPM)
12: #include <petsc/private/deviceimpl.h>
13: #endif
15: /*@C
16: PetscIgnoreErrorHandler - Deprecated, use `PetscReturnErrorHandler()`. Ignores the error, allows program to continue as if error did not occur
18: Not Collective, No Fortran Support
20: Input Parameters:
21: + comm - communicator over which error occurred
22: . line - the line number of the error (indicated by __LINE__)
23: . fun - the function name
24: . file - the file in which the error was detected (indicated by __FILE__)
25: . mess - an error text string, usually just printed to the screen
26: . n - the generic error number
27: . p - specific error number
28: - ctx - error handler context
30: Level: developer
32: Note:
33: Users do not directly call this routine
35: .seealso: `PetscReturnErrorHandler()`
36: @*/
37: PetscErrorCode PetscIgnoreErrorHandler(MPI_Comm comm, int line, const char *fun, const char *file, PetscErrorCode n, PetscErrorType p, const char *mess, void *ctx)
38: {
39: (void)comm;
40: (void)line;
41: (void)fun;
42: (void)file;
43: (void)p;
44: (void)mess;
45: (void)ctx;
46: return n;
47: }
49: /* ---------------------------------------------------------------------------------------*/
51: static char arch[128], hostname[128], username[128], pname[PETSC_MAX_PATH_LEN], date[128];
52: static PetscBool PetscErrorPrintfInitializeCalled = PETSC_FALSE;
53: static char version[256];
55: /*
56: Initializes arch, hostname, username, date so that system calls do NOT need
57: to be made during the error handler.
58: */
59: PetscErrorCode PetscErrorPrintfInitialize(void)
60: {
61: PetscBool use_stdout = PETSC_FALSE, use_none = PETSC_FALSE;
63: PetscFunctionBegin;
64: PetscCall(PetscGetArchType(arch, sizeof(arch)));
65: PetscCall(PetscGetHostName(hostname, sizeof(hostname)));
66: PetscCall(PetscGetUserName(username, sizeof(username)));
67: PetscCall(PetscGetProgramName(pname, sizeof(pname)));
68: PetscCall(PetscGetDate(date, sizeof(date)));
69: PetscCall(PetscGetVersion(version, sizeof(version)));
71: PetscCall(PetscOptionsGetBool(NULL, NULL, "-error_output_stdout", &use_stdout, NULL));
72: if (use_stdout) PETSC_STDERR = PETSC_STDOUT;
73: PetscCall(PetscOptionsGetBool(NULL, NULL, "-error_output_none", &use_none, NULL));
74: if (use_none) PetscErrorPrintf = PetscErrorPrintfNone;
75: PetscErrorPrintfInitializeCalled = PETSC_TRUE;
76: PetscFunctionReturn(PETSC_SUCCESS);
77: }
79: PetscErrorCode PetscErrorPrintfNone(const char format[], ...)
80: {
81: return PETSC_SUCCESS;
82: }
84: PetscErrorCode PetscErrorPrintfDefault(const char format[], ...)
85: {
86: va_list Argp;
87: static PetscBool PetscErrorPrintfCalled = PETSC_FALSE;
89: /*
90: This function does not call PetscFunctionBegin and PetscFunctionReturn() because
91: it may be called by PetscStackView().
93: This function does not do error checking because it is called by the error handlers.
94: */
96: if (!PetscErrorPrintfCalled) {
97: PetscErrorPrintfCalled = PETSC_TRUE;
99: /*
100: On the SGI machines and Cray T3E, if errors are generated "simultaneously" by
101: different processors, the messages are printed all jumbled up; to try to
102: prevent this we have each processor wait based on their rank
103: */
104: #if defined(PETSC_CAN_SLEEP_AFTER_ERROR)
105: {
106: PetscMPIInt rank = PetscGlobalRank > 8 ? 8 : PetscGlobalRank;
107: (void)PetscSleep((PetscReal)rank);
108: }
109: #endif
110: }
112: (void)PetscFPrintf(PETSC_COMM_SELF, PETSC_STDERR, "[%d]PETSC ERROR: ", PetscGlobalRank);
113: va_start(Argp, format);
114: (void)(*PetscVFPrintf)(PETSC_STDERR, format, Argp);
115: va_end(Argp);
116: return PETSC_SUCCESS;
117: }
119: /*
120: On some systems when the stderr is nested through several levels of shell script
121: before being passed to a file the isatty() falsely returns true resulting in
122: the screen highlight variables being passed through the test harness. Therefore
123: simply do not highlight when the PETSC_STDERR is PETSC_STDOUT.
124: */
125: static void PetscErrorPrintfHilight(void)
126: {
127: #if defined(PETSC_HAVE_UNISTD_H) && defined(PETSC_USE_ISATTY)
128: if (PetscErrorPrintf == PetscErrorPrintfDefault && PETSC_STDERR != PETSC_STDOUT) {
129: if (isatty(fileno(PETSC_STDERR))) fprintf(PETSC_STDERR, "\033[1;31m");
130: }
131: #endif
132: }
134: static void PetscErrorPrintfNormal(void)
135: {
136: #if defined(PETSC_HAVE_UNISTD_H) && defined(PETSC_USE_ISATTY)
137: if (PetscErrorPrintf == PetscErrorPrintfDefault && PETSC_STDERR != PETSC_STDOUT) {
138: if (isatty(fileno(PETSC_STDERR))) fprintf(PETSC_STDERR, "\033[0;39m\033[0;49m");
139: }
140: #endif
141: }
143: PETSC_EXTERN PetscErrorCode PetscOptionsViewError(void);
145: static PETSC_TLS PetscBool petsc_traceback_error_silent = PETSC_FALSE;
147: /*@C
149: PetscTraceBackErrorHandler - Default error handler routine that generates
150: a traceback on error detection.
152: Not Collective, No Fortran Support
154: Input Parameters:
155: + comm - communicator over which error occurred
156: . line - the line number of the error (usually indicated by `__LINE__` in the calling routine)
157: . fun - the function name
158: . file - the file in which the error was detected (usually indicated by `__FILE__` in the calling routine)
159: . mess - an error text string, usually just printed to the screen
160: . n - the generic error number
161: . p - `PETSC_ERROR_INITIAL` if this is the first call the error handler, otherwise `PETSC_ERROR_REPEAT`
162: - ctx - error handler context
164: Options Database Keys:
165: + -error_output_stdout - output the error messages to `stdout` instead of the default `stderr`
166: - -error_output_none - do not output the error messages
168: Notes:
169: Users do not directly call this routine
171: Use `PetscPushErrorHandler()` to set the desired error handler.
173: Level: developer
175: .seealso: `PetscError()`, `PetscPushErrorHandler()`, `PetscPopErrorHandler()`, `PetscAttachDebuggerErrorHandler()`,
176: `PetscAbortErrorHandler()`, `PetscMPIAbortErrorHandler()`, `PetscReturnErrorHandler()`, `PetscEmacsClientErrorHandler()`,
177: `PETSC_ERROR_INITIAL`, `PETSC_ERROR_REPEAT`, `PetscErrorCode`, `PetscErrorType`
178: @*/
179: PetscErrorCode PetscTraceBackErrorHandler(MPI_Comm comm, int line, const char *fun, const char *file, PetscErrorCode n, PetscErrorType p, const char *mess, void *ctx)
180: {
181: PetscMPIInt rank = 0;
183: (void)ctx;
184: if (comm != PETSC_COMM_SELF) MPI_Comm_rank(comm, &rank);
186: // reinitialize the error handler when a new initializing error is detected
187: if (p != PETSC_ERROR_REPEAT) {
188: petsc_traceback_error_silent = PETSC_FALSE;
189: if (PetscCIEnabledPortableErrorOutput) {
190: PetscMPIInt size = 1;
192: if (comm != MPI_COMM_NULL) MPI_Comm_size(comm, &size);
193: petscabortmpifinalize = (size == PetscGlobalSize) ? PETSC_TRUE : PETSC_FALSE;
194: }
195: }
197: if (rank == 0 && (!PetscCIEnabledPortableErrorOutput || PetscGlobalRank == 0) && (p != PETSC_ERROR_REPEAT || !petsc_traceback_error_silent)) {
198: static int cnt = 1;
199: PetscBool python = (n == PETSC_ERR_PYTHON && cnt == 1) ? PETSC_TRUE : PETSC_FALSE;
201: if (p == PETSC_ERROR_INITIAL || python) {
202: PetscErrorPrintfHilight();
203: (void)(*PetscErrorPrintf)("--------------------- Error Message --------------------------------------------------------------\n");
204: PetscErrorPrintfNormal();
205: if (cnt > 1) {
206: (void)(*PetscErrorPrintf)(" It appears a new error in the code was triggered after a previous error, possibly because:\n");
207: (void)(*PetscErrorPrintf)(" - The first error was not properly handled via (for example) the use of\n");
208: (void)(*PetscErrorPrintf)(" PetscCall(TheFunctionThatErrors()); or\n");
209: (void)(*PetscErrorPrintf)(" - The second error was triggered while handling the first error.\n");
210: (void)(*PetscErrorPrintf)(" Above is the traceback for the previous unhandled error, below the traceback for the next error\n");
211: (void)(*PetscErrorPrintf)(" ALL ERRORS in the PETSc libraries are fatal, you should add the appropriate error checking to the code\n");
212: cnt = 1;
213: }
214: }
215: if (cnt == 1) {
216: if (n == PETSC_ERR_MEM || n == PETSC_ERR_MEM_LEAK) (void)PetscErrorMemoryMessage(n);
217: else {
218: const char *text;
219: (void)PetscErrorMessage(n, &text, NULL);
220: if (text) (void)(*PetscErrorPrintf)("%s\n", text);
221: }
222: if (python) (void)PetscPythonPrintError();
223: else if (mess) (void)(*PetscErrorPrintf)("%s\n", mess);
224: #if defined(PETSC_PKG_CUDA_MIN_ARCH)
225: int confCudaArch = PETSC_PKG_CUDA_MIN_ARCH; // if PETSc was configured with numbered CUDA arches, get the min arch.
226: int runCudaArch = PetscDeviceCUPMRuntimeArch; // 0 indicates the code has never initialized a cuda device.
227: if (runCudaArch && confCudaArch > runCudaArch) {
228: (void)(*PetscErrorPrintf)("WARNING! Run on a CUDA device with GPU architecture %d, but PETSc was configured with a minimal GPU architecture %d.\n", runCudaArch, confCudaArch);
229: (void)(*PetscErrorPrintf)("If it is a cudaErrorNoKernelImageForDevice error, you may need to reconfigure PETSc with --with-cuda-arch=%d or --with-cuda-arch=%d,%d\n", runCudaArch, runCudaArch, confCudaArch);
230: }
231: #endif
232: (void)PetscOptionsLeftError();
233: (void)(*PetscErrorPrintf)("See https://petsc.org/release/faq/ for trouble shooting.\n");
234: if (!PetscCIEnabledPortableErrorOutput) {
235: size_t clen;
237: (void)(*PetscErrorPrintf)("%s\n", version);
238: if (PetscErrorPrintfInitializeCalled) (void)(*PetscErrorPrintf)("%s with %d MPI process(es) and PETSC_ARCH %s on %s by %s %s\n", pname, PetscGlobalSize, arch, hostname, username, date);
239: (void)PetscStrlen(petscconfigureoptions, &clen);
240: (void)(*PetscErrorPrintf)("Configure options: %s\n", clen ? petscconfigureoptions : "none used");
241: }
242: }
243: /* print line of stack trace */
244: if (fun) (void)(*PetscErrorPrintf)("#%d %s() at %s:%d\n", cnt++, fun, PetscCIFilename(file), PetscCILinenumber(line));
245: else if (file) (void)(*PetscErrorPrintf)("#%d %s:%d\n", cnt++, PetscCIFilename(file), PetscCILinenumber(line));
246: if (fun) {
247: PetscBool ismain = PETSC_FALSE;
249: (void)PetscStrncmp(fun, "main", 4, &ismain);
250: if (ismain) {
251: if ((n <= PETSC_ERR_MIN_VALUE) || (n >= PETSC_ERR_MAX_VALUE)) (void)(*PetscErrorPrintf)("Reached the main program with an out-of-range error code %d. This should never happen\n", n);
252: (void)PetscOptionsViewError();
253: PetscErrorPrintfHilight();
254: (void)(*PetscErrorPrintf)("----------------End of Error Message -------send entire error message to petsc-maint@mcs.anl.gov----------\n");
255: PetscErrorPrintfNormal();
256: }
257: }
258: } else {
259: // silence this process's stacktrace if it is not the root of an originating error
260: if (p != PETSC_ERROR_REPEAT && rank) petsc_traceback_error_silent = PETSC_TRUE;
261: if (fun) {
262: PetscBool ismain = PETSC_FALSE;
264: (void)PetscStrncmp(fun, "main", 4, &ismain);
265: if (ismain && petsc_traceback_error_silent) {
266: /* This results from PetscError() being called in main: PETSCABORT()
267: will be called after the error handler. But this thread is not the
268: root rank of the communicator that initialized the error. So sleep
269: to allow the root thread to finish its printing.
271: (Unless this is running CI, in which case do not sleep because
272: we expect all processes to call MPI_Finalize() and make a clean
273: exit.) */
274: if (!PetscCIEnabledPortableErrorOutput) (void)PetscSleep(10.0);
275: }
276: }
277: }
278: return n;
279: }