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: static char arch[128], hostname[128], username[128], pname[PETSC_MAX_PATH_LEN], date[128];
16: static PetscBool PetscErrorPrintfInitializeCalled = PETSC_FALSE;
17: static char version[256];
19: /*
20: Initializes arch, hostname, username, date so that system calls do NOT need
21: to be made during the error handler.
22: */
23: PetscErrorCode PetscErrorPrintfInitialize(void)
24: {
25: PetscBool use_stdout = PETSC_FALSE, use_none = PETSC_FALSE;
27: PetscFunctionBegin;
28: PetscCall(PetscGetArchType(arch, sizeof(arch)));
29: PetscCall(PetscGetHostName(hostname, sizeof(hostname)));
30: PetscCall(PetscGetUserName(username, sizeof(username)));
31: PetscCall(PetscGetProgramName(pname, sizeof(pname)));
32: PetscCall(PetscGetDate(date, sizeof(date)));
33: PetscCall(PetscGetVersion(version, sizeof(version)));
35: PetscCall(PetscOptionsGetBool(NULL, NULL, "-error_output_stdout", &use_stdout, NULL));
36: if (use_stdout) PETSC_STDERR = PETSC_STDOUT;
37: PetscCall(PetscOptionsGetBool(NULL, NULL, "-error_output_none", &use_none, NULL));
38: if (use_none) PetscErrorPrintf = PetscErrorPrintfNone;
39: PetscErrorPrintfInitializeCalled = PETSC_TRUE;
40: PetscFunctionReturn(PETSC_SUCCESS);
41: }
43: PetscErrorCode PetscErrorPrintfNone(const char format[], ...)
44: {
45: return PETSC_SUCCESS;
46: }
48: PetscErrorCode PetscErrorPrintfDefault(const char format[], ...)
49: {
50: va_list Argp;
51: static PetscBool PetscErrorPrintfCalled = PETSC_FALSE;
53: /*
54: This function does not call PetscFunctionBegin and PetscFunctionReturn() because
55: it may be called by PetscStackView().
57: This function does not do error checking because it is called by the error handlers.
58: */
60: if (!PetscErrorPrintfCalled) {
61: PetscErrorPrintfCalled = PETSC_TRUE;
63: /*
64: On the SGI machines and Cray T3E, if errors are generated "simultaneously" by
65: different processors, the messages are printed all jumbled up; to try to
66: prevent this we have each processor wait based on their rank
67: */
68: #if defined(PETSC_CAN_SLEEP_AFTER_ERROR)
69: {
70: PetscMPIInt rank = PetscGlobalRank > 8 ? 8 : PetscGlobalRank;
71: (void)PetscSleep((PetscReal)rank);
72: }
73: #endif
74: }
76: (void)PetscFPrintf(PETSC_COMM_SELF, PETSC_STDERR, "[%d]PETSC ERROR: ", PetscGlobalRank);
77: va_start(Argp, format);
78: (void)(*PetscVFPrintf)(PETSC_STDERR, format, Argp);
79: va_end(Argp);
80: return PETSC_SUCCESS;
81: }
83: /*
84: On some systems when the stderr is nested through several levels of shell script
85: before being passed to a file the isatty() falsely returns true resulting in
86: the screen highlight variables being passed through the test harness. Therefore
87: simply do not highlight when the PETSC_STDERR is PETSC_STDOUT.
88: */
89: static void PetscErrorPrintfHilight(void)
90: {
91: #if defined(PETSC_HAVE_UNISTD_H) && defined(PETSC_USE_ISATTY)
92: if (PetscErrorPrintf == PetscErrorPrintfDefault && PETSC_STDERR != PETSC_STDOUT) {
93: if (isatty(fileno(PETSC_STDERR))) fprintf(PETSC_STDERR, "\033[1;31m");
94: }
95: #endif
96: }
98: static void PetscErrorPrintfNormal(void)
99: {
100: #if defined(PETSC_HAVE_UNISTD_H) && defined(PETSC_USE_ISATTY)
101: if (PetscErrorPrintf == PetscErrorPrintfDefault && PETSC_STDERR != PETSC_STDOUT) {
102: if (isatty(fileno(PETSC_STDERR))) fprintf(PETSC_STDERR, "\033[0;39m\033[0;49m");
103: }
104: #endif
105: }
107: PETSC_EXTERN PetscErrorCode PetscOptionsViewError(void);
109: static PETSC_TLS PetscBool petsc_traceback_error_silent = PETSC_FALSE;
111: /*@C
112: PetscTraceBackErrorHandler - Default error handler routine that generates a traceback on error detection.
114: Not Collective, No Fortran Support
116: Input Parameters:
117: + comm - communicator over which error occurred
118: . line - the line number of the error (usually indicated by `__LINE__` in the calling routine)
119: . fun - the function name
120: . file - the file in which the error was detected (usually indicated by `__FILE__` in the calling routine)
121: . mess - an error text string, usually just printed to the screen
122: . n - the generic error number
123: . p - `PETSC_ERROR_INITIAL` if this is the first call the error handler, otherwise `PETSC_ERROR_REPEAT`
124: - ctx - error handler context
126: Options Database Keys:
127: + -error_output_stdout - output the error messages to `stdout` instead of the default `stderr`
128: - -error_output_none - do not output the error messages
130: Notes:
131: Users do not directly call this routine
133: Use `PetscPushErrorHandler()` to set the desired error handler.
135: Level: developer
137: .seealso: `PetscError()`, `PetscPushErrorHandler()`, `PetscPopErrorHandler()`, `PetscAttachDebuggerErrorHandler()`,
138: `PetscAbortErrorHandler()`, `PetscMPIAbortErrorHandler()`, `PetscReturnErrorHandler()`, `PetscEmacsClientErrorHandler()`,
139: `PETSC_ERROR_INITIAL`, `PETSC_ERROR_REPEAT`, `PetscErrorCode`, `PetscErrorType`
140: @*/
141: PetscErrorCode PetscTraceBackErrorHandler(MPI_Comm comm, int line, const char *fun, const char *file, PetscErrorCode n, PetscErrorType p, const char *mess, PetscCtx ctx)
142: {
143: PetscMPIInt rank = 0;
145: (void)ctx;
146: if (comm != PETSC_COMM_SELF) MPI_Comm_rank(comm, &rank);
148: // reinitialize the error handler when a new initializing error is detected
149: if (p != PETSC_ERROR_REPEAT) {
150: petsc_traceback_error_silent = PETSC_FALSE;
151: if (PetscCIEnabledPortableErrorOutput) {
152: PetscMPIInt size = 1;
154: if (comm != MPI_COMM_NULL) MPI_Comm_size(comm, &size);
155: petscabortmpifinalize = (size == PetscGlobalSize) ? PETSC_TRUE : PETSC_FALSE;
156: }
157: }
159: if (rank == 0 && (!PetscCIEnabledPortableErrorOutput || PetscGlobalRank == 0) && (p != PETSC_ERROR_REPEAT || !petsc_traceback_error_silent)) {
160: static int cnt = 1;
161: PetscBool python = (n == PETSC_ERR_PYTHON && cnt == 1) ? PETSC_TRUE : PETSC_FALSE;
163: if (p == PETSC_ERROR_INITIAL || python) {
164: PetscErrorPrintfHilight();
165: (void)(*PetscErrorPrintf)("--------------------- Error Message --------------------------------------------------------------\n");
166: PetscErrorPrintfNormal();
167: if (cnt > 1) {
168: (void)(*PetscErrorPrintf)(" It appears a new error in the code was triggered after a previous error, possibly because:\n");
169: (void)(*PetscErrorPrintf)(" - The first error was not properly handled via (for example) the use of\n");
170: (void)(*PetscErrorPrintf)(" PetscCall(TheFunctionThatErrors()); or\n");
171: (void)(*PetscErrorPrintf)(" - The second error was triggered while handling the first error.\n");
172: (void)(*PetscErrorPrintf)(" Above is the traceback for the previous unhandled error, below the traceback for the next error\n");
173: (void)(*PetscErrorPrintf)(" ALL ERRORS in the PETSc libraries are fatal, you should add the appropriate error checking to the code\n");
174: cnt = 1;
175: }
176: }
177: if (cnt == 1) {
178: if (n == PETSC_ERR_MEM || n == PETSC_ERR_MEM_LEAK) (void)PetscErrorMemoryMessage(n);
179: else {
180: const char *text;
181: (void)PetscErrorMessage(n, &text, NULL);
182: if (text) (void)(*PetscErrorPrintf)("%s\n", text);
183: }
184: if (python) (void)PetscPythonPrintError();
185: else if (mess) (void)(*PetscErrorPrintf)("%s\n", mess);
186: #if defined(PETSC_PKG_CUDA_MIN_ARCH)
187: int confCudaArch = PETSC_PKG_CUDA_MIN_ARCH; // if PETSc was configured with numbered CUDA arches, get the min arch.
188: int runCudaArch = PetscDeviceCUPMRuntimeArch; // 0 indicates the code has never initialized a cuda device.
189: if (runCudaArch && confCudaArch > runCudaArch) {
190: (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);
191: (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);
192: }
193: #endif
194: (void)PetscOptionsLeftError();
195: (void)(*PetscErrorPrintf)("See https://petsc.org/release/faq/ for trouble shooting.\n");
196: if (!PetscCIEnabledPortableErrorOutput) {
197: size_t clen;
199: (void)(*PetscErrorPrintf)("%s\n", version);
200: 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);
201: (void)PetscStrlen(petscconfigureoptions, &clen);
202: (void)(*PetscErrorPrintf)("Configure options: %s\n", clen ? petscconfigureoptions : "none used");
203: }
204: }
205: /* print line of stack trace */
206: if (fun) (void)(*PetscErrorPrintf)("#%d %s() at %s:%d\n", cnt++, fun, PetscCIFilename(file), PetscCILinenumber(line));
207: else if (file) (void)(*PetscErrorPrintf)("#%d %s:%d\n", cnt++, PetscCIFilename(file), PetscCILinenumber(line));
208: if (fun) {
209: PetscBool ismain = PETSC_FALSE;
211: (void)PetscStrncmp(fun, "main", 4, &ismain);
212: if (ismain) {
213: 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);
214: (void)PetscOptionsViewError();
215: PetscErrorPrintfHilight();
216: (void)(*PetscErrorPrintf)("----------------End of Error Message -------send entire error message to petsc-maint@mcs.anl.gov----------\n");
217: PetscErrorPrintfNormal();
218: }
219: }
220: } else {
221: // silence this process's stacktrace if it is not the root of an originating error
222: if (p != PETSC_ERROR_REPEAT && rank) petsc_traceback_error_silent = PETSC_TRUE;
223: if (fun) {
224: PetscBool ismain = PETSC_FALSE;
226: (void)PetscStrncmp(fun, "main", 4, &ismain);
227: if (ismain && petsc_traceback_error_silent) {
228: /* This results from PetscError() being called in main: PETSCABORT()
229: will be called after the error handler. But this thread is not the
230: root rank of the communicator that initialized the error. So sleep
231: to allow the root thread to finish its printing.
233: (Unless this is running CI, in which case do not sleep because
234: we expect all processes to call MPI_Finalize() and make a clean
235: exit.) */
236: if (!PetscCIEnabledPortableErrorOutput) (void)PetscSleep(10.0);
237: }
238: }
239: }
240: return n;
241: }