/* T E M P L A T E P A R S E R ============================= by Jerzy.Borkowski@obs.unige.ch Integral Science Data Center ch. d'Ecogia 16 1290 Versoix Switzerland 14-Oct-98: initial release 16-Oct-98: code cleanup, #include included, now gcc -Wall prints no warnings during compilation. Bugfix: now one can specify additional columns in group HDU. Autoindexing also works in this situation (colunms are number from 7 however). 17-Oct-98: bugfix: complex keywords were incorrectly written (was TCOMPLEX should be TDBLCOMPLEX). 20-Oct-98: bugfix: parser was writing EXTNAME twice, when first HDU in template is defined with XTENSION IMAGE then parser creates now dummy PHDU, SIMPLE T is now allowed only at most once and in first HDU only. WARNING: one should not define EXTNAME keyword for GROUP HDUs, as they have them already defined by parser (EXTNAME = GROUPING). Parser accepts EXTNAME oin GROUP HDU definition, but in this case multiple EXTNAME keywords will present in HDU header. 23-Oct-98: bugfix: unnecessary space was written to FITS file for blank keywords. 24-Oct-98: syntax change: empty lines and lines with only whitespaces are written to FITS files as blank keywords (if inside group/hdu definition). Previously lines had to have at least 8 spaces. Please note, that due to pecularities of CFITSIO if the last keyword(s) defined for given HDU are blank keywords consisting of only 80 spaces, then (some of) those keywords may be silently deleted by CFITSIO. 13-Nov-98: bugfix: parser was writing GRPNAME twice. Parser still creates GRPNAME keywords for GROUP HDU's which do not specify them. However, values (of form DEFAULT_GROUP_XXX) are assigned not necessarily in order HDUs appear in template file, but rather in order parser completes their creation in FITS file. Also, when including files, if fopen fails, parser tries to open file with a name = directory_of_top_level file + name of file to be included, as long as name of file to be included does not specify absolute pathname. 16-Nov-98: bugfix to bugfix from 13-Nov-98 19-Nov-98: EXTVER keyword is now automatically assigned value by parser. 17-Dev-98: 2 new things added: 1st: CFITSIO_INCLUDE_FILES environment variable can contain a colon separated list of directories to look for when looking for template include files (and master template also). 2nd: it is now possible to append template to nonempty FITS. file. fitsfile *ff no longer needs to point to an empty FITS file with 0 HDUs in it. All data written by parser will simple be appended at the end of file. 22-Jan-99: changes to parser: when in append mode parser initially scans all existing HDUs to built a list of already used EXTNAME/EXTVERs 22-Jan-99: Bruce O'Neel, bugfix : TLONG should always reference long type variable on OSF/Alpha and on 64-bit archs in general 20-Jun-2002 Wm Pence, added support for the HIERARCH keyword convention in which keyword names can effectively be longer than 8 characters. Example: HIERARCH LongKeywordName = 'value' / comment 30-Jan-2003 Wm Pence, bugfix: ngp_read_xtension was testing for "ASCIITABLE" instead of "TABLE" as the XTENSION value of an ASCII table, and it did not allow for optional trailing spaces in the "IMAGE" or "TABLE" string. 16-Dec-2003 James Peachey: ngp_keyword_all_write was modified to apply comments from the template file to the output file in the case of reserved keywords (e.g. tform#, ttype# etcetera). */ #include #include #ifdef sparc #include #include #endif #include #include "fitsio2.h" #include "grparser.h" NGP_RAW_LINE ngp_curline = { NULL, NULL, NULL, NGP_TTYPE_UNKNOWN, NULL, NGP_FORMAT_OK, 0 }; NGP_RAW_LINE ngp_prevline = { NULL, NULL, NULL, NGP_TTYPE_UNKNOWN, NULL, NGP_FORMAT_OK, 0 }; int ngp_inclevel = 0; /* number of included files, 1 - means mean file */ int ngp_grplevel = 0; /* group nesting level, 0 - means no grouping */ FILE *ngp_fp[NGP_MAX_INCLUDE]; /* stack of included file handles */ int ngp_keyidx = NGP_TOKEN_UNKNOWN; /* index of token in current line */ NGP_TOKEN ngp_linkey; /* keyword after line analyze */ char ngp_master_dir[NGP_MAX_FNAME]; /* directory of top level include file */ NGP_TKDEF ngp_tkdef[] = /* tokens recognized by parser */ { { "\\INCLUDE", NGP_TOKEN_INCLUDE }, { "\\GROUP", NGP_TOKEN_GROUP }, { "\\END", NGP_TOKEN_END }, { "XTENSION", NGP_TOKEN_XTENSION }, { "SIMPLE", NGP_TOKEN_SIMPLE }, { NULL, NGP_TOKEN_UNKNOWN } }; int master_grp_idx = 1; /* current unnamed group in object */ int ngp_extver_tab_size = 0; NGP_EXTVER_TAB *ngp_extver_tab = NULL; int ngp_get_extver(char *extname, int *version) { NGP_EXTVER_TAB *p; char *p2; int i; if ((NULL == extname) || (NULL == version)) return(NGP_BAD_ARG); if ((NULL == ngp_extver_tab) && (ngp_extver_tab_size > 0)) return(NGP_BAD_ARG); if ((NULL != ngp_extver_tab) && (ngp_extver_tab_size <= 0)) return(NGP_BAD_ARG); for (i=0; i 0)) return(NGP_BAD_ARG); if ((NULL != ngp_extver_tab) && (ngp_extver_tab_size <= 0)) return(NGP_BAD_ARG); for (i=0; i ngp_extver_tab[i].version) ngp_extver_tab[i].version = version; return(NGP_OK); } } if (NULL == ngp_extver_tab) { p = (NGP_EXTVER_TAB *)ngp_alloc(sizeof(NGP_EXTVER_TAB)); } else { p = (NGP_EXTVER_TAB *)ngp_realloc(ngp_extver_tab, (ngp_extver_tab_size + 1) * sizeof(NGP_EXTVER_TAB)); } if (NULL == p) return(NGP_NO_MEMORY); p2 = ngp_alloc(strlen(extname) + 1); if (NULL == p2) { ngp_free(p); return(NGP_NO_MEMORY); } strcpy(p2, extname); ngp_extver_tab = p; ngp_extver_tab[ngp_extver_tab_size].extname = p2; ngp_extver_tab[ngp_extver_tab_size].version = version; ngp_extver_tab_size++; return(NGP_OK); } int ngp_delete_extver_tab(void) { int i; if ((NULL == ngp_extver_tab) && (ngp_extver_tab_size > 0)) return(NGP_BAD_ARG); if ((NULL != ngp_extver_tab) && (ngp_extver_tab_size <= 0)) return(NGP_BAD_ARG); if ((NULL == ngp_extver_tab) && (0 == ngp_extver_tab_size)) return(NGP_OK); for (i=0; i= 'a') && (c1 <= 'z')) c1 += ('A' - 'a'); c2 = *p2; if ((c2 >= 'a') && (c2 <= 'z')) c2 += ('A' - 'a'); if (c1 < c2) return(-1); if (c1 > c2) return(1); if (0 == c1) return(0); p1++; p2++; } } int ngp_strcasencmp(char *p1, char *p2, int n) { char c1, c2; int ii; for (ii=0;ii= 'a') && (c1 <= 'z')) c1 += ('A' - 'a'); c2 = *p2; if ((c2 >= 'a') && (c2 <= 'z')) c2 += ('A' - 'a'); if (c1 < c2) return(-1); if (c1 > c2) return(1); if (0 == c1) return(0); p1++; p2++; } return(0); } /* read one line from file */ int ngp_line_from_file(FILE *fp, char **p) { int c, r, llen, allocsize, alen; char *p2; if (NULL == fp) return(NGP_NUL_PTR); /* check for stupid args */ if (NULL == p) return(NGP_NUL_PTR); /* more foolproof checks */ r = NGP_OK; /* initialize stuff, reset err code */ llen = 0; /* 0 characters read so far */ *p = (char *)ngp_alloc(1); /* preallocate 1 byte */ allocsize = 1; /* signal that we have allocated 1 byte */ if (NULL == *p) return(NGP_NO_MEMORY); /* if this failed, system is in dire straits */ for (;;) { c = getc(fp); /* get next character */ if (EOF == c) /* EOF signalled ? */ { if (ferror(fp)) r = NGP_READ_ERR; /* was it real error or simply EOF ? */ if (0 == llen) return(NGP_EOF); /* signal EOF only if 0 characters read so far */ break; } if ('\n' == c) break; /* end of line character ? */ llen++; /* we have new character, make room for it */ alen = ((llen + NGP_ALLOCCHUNK) / NGP_ALLOCCHUNK) * NGP_ALLOCCHUNK; if (alen > allocsize) { p2 = (char *)ngp_realloc(*p, alen); /* realloc buffer, if there is need */ if (NULL == p2) { r = NGP_NO_MEMORY; break; } *p = p2; allocsize = alen; } (*p)[llen - 1] = c; /* copy character to buffer */ } llen++; /* place for terminating \0 */ if (llen != allocsize) { p2 = (char *)ngp_realloc(*p, llen); if (NULL == p2) r = NGP_NO_MEMORY; else { *p = p2; (*p)[llen - 1] = 0; /* copy \0 to buffer */ } } else { (*p)[llen - 1] = 0; /* necessary when line read was empty */ } if ((NGP_EOF != r) && (NGP_OK != r)) /* in case of errors free resources */ { ngp_free(*p); *p = NULL; } return(r); /* return status code */ } /* free current line structure */ int ngp_free_line(void) { if (NULL != ngp_curline.line) { ngp_free(ngp_curline.line); ngp_curline.line = NULL; ngp_curline.name = NULL; ngp_curline.value = NULL; ngp_curline.comment = NULL; ngp_curline.type = NGP_TTYPE_UNKNOWN; ngp_curline.format = NGP_FORMAT_OK; ngp_curline.flags = 0; } return(NGP_OK); } /* free cached line structure */ int ngp_free_prevline(void) { if (NULL != ngp_prevline.line) { ngp_free(ngp_prevline.line); ngp_prevline.line = NULL; ngp_prevline.name = NULL; ngp_prevline.value = NULL; ngp_prevline.comment = NULL; ngp_prevline.type = NGP_TTYPE_UNKNOWN; ngp_prevline.format = NGP_FORMAT_OK; ngp_prevline.flags = 0; } return(NGP_OK); } /* read one line */ int ngp_read_line_buffered(FILE *fp) { ngp_free_line(); /* first free current line (if any) */ if (NULL != ngp_prevline.line) /* if cached, return cached line */ { ngp_curline = ngp_prevline; ngp_prevline.line = NULL; ngp_prevline.name = NULL; ngp_prevline.value = NULL; ngp_prevline.comment = NULL; ngp_prevline.type = NGP_TTYPE_UNKNOWN; ngp_prevline.format = NGP_FORMAT_OK; ngp_prevline.flags = 0; ngp_curline.flags = NGP_LINE_REREAD; return(NGP_OK); } ngp_curline.flags = 0; /* if not cached really read line from file */ return(ngp_line_from_file(fp, &(ngp_curline.line))); } /* unread line */ int ngp_unread_line(void) { if (NULL == ngp_curline.line) /* nothing to unread */ return(NGP_EMPTY_CURLINE); if (NULL != ngp_prevline.line) /* we cannot unread line twice */ return(NGP_UNREAD_QUEUE_FULL); ngp_prevline = ngp_curline; ngp_curline.line = NULL; return(NGP_OK); } /* a first guess line decomposition */ int ngp_extract_tokens(NGP_RAW_LINE *cl) { char *p, *s; int cl_flags, i; p = cl->line; /* start from beginning of line */ if (NULL == p) return(NGP_NUL_PTR); cl->name = cl->value = cl->comment = NULL; cl->type = NGP_TTYPE_UNKNOWN; cl->format = NGP_FORMAT_OK; cl_flags = 0; for (i=0;; i++) /* if 8 spaces at beginning then line is comment */ { if ((0 == *p) || ('\n' == *p)) { /* if line has only blanks -> write blank keyword */ cl->line[0] = 0; /* create empty name (0 length string) */ cl->comment = cl->name = cl->line; cl->type = NGP_TTYPE_RAW; /* signal write unformatted to FITS file */ return(NGP_OK); } if ((' ' != *p) && ('\t' != *p)) break; if (i >= 7) { cl->comment = p + 1; for (s = cl->comment;; s++) /* filter out any EOS characters in comment */ { if ('\n' == *s) *s = 0; if (0 == *s) break; } cl->line[0] = 0; /* create empty name (0 length string) */ cl->name = cl->line; cl->type = NGP_TTYPE_RAW; return(NGP_OK); } p++; } cl->name = p; for (;;) /* we need to find 1st whitespace */ { if ((0 == *p) || ('\n' == *p)) { *p = 0; break; } /* from Richard Mathar, 2002-05-03, add 10 lines: if upper/lowercase HIERARCH followed also by an equal sign... */ if( strncasecmp("HIERARCH",p,strlen("HIERARCH")) == 0 ) { char * const eqsi=strchr(p,'=') ; if( eqsi ) { cl_flags |= NGP_FOUND_EQUAL_SIGN ; p=eqsi ; break ; } } if ((' ' == *p) || ('\t' == *p)) break; if ('=' == *p) { cl_flags |= NGP_FOUND_EQUAL_SIGN; break; } p++; } if (*p) *(p++) = 0; /* found end of keyname so terminate string with zero */ if ((!ngp_strcasecmp("HISTORY", cl->name)) || (!ngp_strcasecmp("COMMENT", cl->name)) || (!ngp_strcasecmp("CONTINUE", cl->name))) { cl->comment = p; for (s = cl->comment;; s++) /* filter out any EOS characters in comment */ { if ('\n' == *s) *s = 0; if (0 == *s) break; } cl->type = NGP_TTYPE_RAW; return(NGP_OK); } if (!ngp_strcasecmp("\\INCLUDE", cl->name)) { for (;; p++) if ((' ' != *p) && ('\t' != *p)) break; /* skip whitespace */ cl->value = p; for (s = cl->value;; s++) /* filter out any EOS characters */ { if ('\n' == *s) *s = 0; if (0 == *s) break; } cl->type = NGP_TTYPE_UNKNOWN; return(NGP_OK); } for (;; p++) { if ((0 == *p) || ('\n' == *p)) return(NGP_OK); /* test if at end of string */ if ((' ' == *p) || ('\t' == *p)) continue; /* skip whitespace */ if (cl_flags & NGP_FOUND_EQUAL_SIGN) break; if ('=' != *p) break; /* ignore initial equal sign */ cl_flags |= NGP_FOUND_EQUAL_SIGN; } if ('/' == *p) /* no value specified, comment only */ { p++; if ((' ' == *p) || ('\t' == *p)) p++; cl->comment = p; for (s = cl->comment;; s++) /* filter out any EOS characters in comment */ { if ('\n' == *s) *s = 0; if (0 == *s) break; } return(NGP_OK); } if ('\'' == *p) /* we have found string within quotes */ { cl->value = s = ++p; /* set pointer to beginning of that string */ cl->type = NGP_TTYPE_STRING; /* signal that it is of string type */ for (;;) /* analyze it */ { if ((0 == *p) || ('\n' == *p)) /* end of line -> end of string */ { *s = 0; return(NGP_OK); } if ('\'' == *p) /* we have found doublequote */ { if ((0 == p[1]) || ('\n' == p[1]))/* doublequote is the last character in line */ { *s = 0; return(NGP_OK); } if (('\t' == p[1]) || (' ' == p[1])) /* duoblequote was string terminator */ { *s = 0; p++; break; } if ('\'' == p[1]) p++; /* doublequote is inside string, convert "" -> " */ } *(s++) = *(p++); /* compact string in place, necess. by "" -> " conversion */ } } else /* regular token */ { cl->value = p; /* set pointer to token */ cl->type = NGP_TTYPE_UNKNOWN; /* we dont know type at the moment */ for (;; p++) /* we need to find 1st whitespace */ { if ((0 == *p) || ('\n' == *p)) { *p = 0; return(NGP_OK); } if ((' ' == *p) || ('\t' == *p)) break; } if (*p) *(p++) = 0; /* found so terminate string with zero */ } for (;; p++) { if ((0 == *p) || ('\n' == *p)) return(NGP_OK); /* test if at end of string */ if ((' ' != *p) && ('\t' != *p)) break; /* skip whitespace */ } if ('/' == *p) /* no value specified, comment only */ { p++; if ((' ' == *p) || ('\t' == *p)) p++; cl->comment = p; for (s = cl->comment;; s++) /* filter out any EOS characters in comment */ { if ('\n' == *s) *s = 0; if (0 == *s) break; } return(NGP_OK); } cl->format = NGP_FORMAT_ERROR; return(NGP_OK); /* too many tokens ... */ } /* try to open include file. If open fails and fname does not specify absolute pathname, try to open fname in any directory specified in CFITSIO_INCLUDE_FILES environment variable. Finally try to open fname relative to ngp_master_dir, which is directory of top level include file */ int ngp_include_file(char *fname) /* try to open include file */ { char *p, *p2, *cp, *envar, envfiles[NGP_MAX_ENVFILES]; if (NULL == fname) return(NGP_NUL_PTR); if (ngp_inclevel >= NGP_MAX_INCLUDE) /* too many include files */ return(NGP_INC_NESTING); if (NULL == (ngp_fp[ngp_inclevel] = fopen(fname, "r"))) { /* if simple open failed .. */ envar = getenv("CFITSIO_INCLUDE_FILES"); /* scan env. variable, and retry to open */ if (NULL != envar) /* is env. variable defined ? */ { strncpy(envfiles, envar, NGP_MAX_ENVFILES - 1); envfiles[NGP_MAX_ENVFILES - 1] = 0; /* copy search path to local variable, env. is fragile */ for (p2 = strtok(envfiles, ":"); NULL != p2; p2 = strtok(NULL, ":")) { cp = (char *)ngp_alloc(strlen(fname) + strlen(p2) + 2); if (NULL == cp) return(NGP_NO_MEMORY); strcpy(cp, p2); #ifdef MSDOS strcat(cp, "\\"); /* abs. pathname for MSDOS */ #else strcat(cp, "/"); /* and for unix */ #endif strcat(cp, fname); ngp_fp[ngp_inclevel] = fopen(cp, "r"); ngp_free(cp); if (NULL != ngp_fp[ngp_inclevel]) break; } } if (NULL == ngp_fp[ngp_inclevel]) /* finally try to open relative to top level */ { #ifdef MSDOS if ('\\' == fname[0]) return(NGP_ERR_FOPEN); /* abs. pathname for MSDOS, does not support C:\\PATH */ #else if ('/' == fname[0]) return(NGP_ERR_FOPEN); /* and for unix */ #endif if (0 == ngp_master_dir[0]) return(NGP_ERR_FOPEN); p = ngp_alloc(strlen(fname) + strlen(ngp_master_dir) + 1); if (NULL == p) return(NGP_NO_MEMORY); strcpy(p, ngp_master_dir); /* construct composite pathname */ strcat(p, fname); /* comp = master + fname */ ngp_fp[ngp_inclevel] = fopen(p, "r");/* try to open composite */ ngp_free(p); /* we don't need buffer anymore */ if (NULL == ngp_fp[ngp_inclevel]) return(NGP_ERR_FOPEN); /* fail if error */ } } ngp_inclevel++; return(NGP_OK); } /* read line in the intelligent way. All \INCLUDE directives are handled, empty and comment line skipped. If this function returns NGP_OK, than decomposed line (name, type, value in proper type and comment) are stored in ngp_linkey structure. ignore_blank_lines parameter is zero when parser is inside GROUP or HDU definition. Nonzero otherwise. */ int ngp_read_line(int ignore_blank_lines) { int r, nc; unsigned k; if (ngp_inclevel <= 0) /* do some sanity checking first */ { ngp_keyidx = NGP_TOKEN_EOF; /* no parents, so report error */ return(NGP_OK); } if (ngp_inclevel > NGP_MAX_INCLUDE) return(NGP_INC_NESTING); if (NULL == ngp_fp[ngp_inclevel - 1]) return(NGP_NUL_PTR); for (;;) { switch (r = ngp_read_line_buffered(ngp_fp[ngp_inclevel - 1])) { case NGP_EOF: ngp_inclevel--; /* end of file, revert to parent */ if (ngp_fp[ngp_inclevel]) /* we can close old file */ fclose(ngp_fp[ngp_inclevel]); ngp_fp[ngp_inclevel] = NULL; if (ngp_inclevel <= 0) { ngp_keyidx = NGP_TOKEN_EOF; /* no parents, so report error */ return(NGP_OK); } continue; case NGP_OK: if (ngp_curline.flags & NGP_LINE_REREAD) return(r); break; default: return(r); } switch (ngp_curline.line[0]) { case 0: if (0 == ignore_blank_lines) break; /* ignore empty lines if told so */ case '#': continue; /* ignore comment lines */ } r = ngp_extract_tokens(&ngp_curline); /* analyse line, extract tokens and comment */ if (NGP_OK != r) return(r); if (NULL == ngp_curline.name) continue; /* skip lines consisting only of whitespaces */ for (k = 0; k < strlen(ngp_curline.name); k++) { if ((ngp_curline.name[k] >= 'a') && (ngp_curline.name[k] <= 'z')) ngp_curline.name[k] += 'A' - 'a'; /* force keyword to be upper case */ if (k == 7) break; /* only first 8 chars are required to be upper case */ } for (k=0;; k++) /* find index of keyword in keyword table */ { if (NGP_TOKEN_UNKNOWN == ngp_tkdef[k].code) break; if (0 == strcmp(ngp_curline.name, ngp_tkdef[k].name)) break; } ngp_keyidx = ngp_tkdef[k].code; /* save this index, grammar parser will need this */ if (NGP_TOKEN_INCLUDE == ngp_keyidx) /* if this is \INCLUDE keyword, try to include file */ { if (NGP_OK != (r = ngp_include_file(ngp_curline.value))) return(r); continue; /* and read next line */ } ngp_linkey.type = NGP_TTYPE_UNKNOWN; /* now, get the keyword type, it's a long story ... */ if (NULL != ngp_curline.value) /* if no value given signal it */ { if (NGP_TTYPE_STRING == ngp_curline.type) /* string type test */ { ngp_linkey.type = NGP_TTYPE_STRING; ngp_linkey.value.s = ngp_curline.value; } if (NGP_TTYPE_UNKNOWN == ngp_linkey.type) /* bool type test */ { if ((!ngp_strcasecmp("T", ngp_curline.value)) || (!ngp_strcasecmp("F", ngp_curline.value))) { ngp_linkey.type = NGP_TTYPE_BOOL; ngp_linkey.value.b = (ngp_strcasecmp("T", ngp_curline.value) ? 0 : 1); } } if (NGP_TTYPE_UNKNOWN == ngp_linkey.type) /* complex type test */ { if (2 == sscanf(ngp_curline.value, "(%lg,%lg)%n", &(ngp_linkey.value.c.re), &(ngp_linkey.value.c.im), &nc)) { if ((' ' == ngp_curline.value[nc]) || ('\t' == ngp_curline.value[nc]) || ('\n' == ngp_curline.value[nc]) || (0 == ngp_curline.value[nc])) { ngp_linkey.type = NGP_TTYPE_COMPLEX; } } } if (NGP_TTYPE_UNKNOWN == ngp_linkey.type) /* real type test */ { if (strchr(ngp_curline.value, '.') && (1 == sscanf(ngp_curline.value, "%lg%n", &(ngp_linkey.value.d), &nc))) { if ((' ' == ngp_curline.value[nc]) || ('\t' == ngp_curline.value[nc]) || ('\n' == ngp_curline.value[nc]) || (0 == ngp_curline.value[nc])) { ngp_linkey.type = NGP_TTYPE_REAL; } } } if (NGP_TTYPE_UNKNOWN == ngp_linkey.type) /* integer type test */ { if (1 == sscanf(ngp_curline.value, "%d%n", &(ngp_linkey.value.i), &nc)) { if ((' ' == ngp_curline.value[nc]) || ('\t' == ngp_curline.value[nc]) || ('\n' == ngp_curline.value[nc]) || (0 == ngp_curline.value[nc])) { ngp_linkey.type = NGP_TTYPE_INT; } } } if (NGP_TTYPE_UNKNOWN == ngp_linkey.type) /* force string type */ { ngp_linkey.type = NGP_TTYPE_STRING; ngp_linkey.value.s = ngp_curline.value; } } else { if (NGP_TTYPE_RAW == ngp_curline.type) ngp_linkey.type = NGP_TTYPE_RAW; else ngp_linkey.type = NGP_TTYPE_NULL; } if (NULL != ngp_curline.comment) { strncpy(ngp_linkey.comment, ngp_curline.comment, NGP_MAX_COMMENT); /* store comment */ ngp_linkey.comment[NGP_MAX_COMMENT - 1] = 0; } else { ngp_linkey.comment[0] = 0; } strncpy(ngp_linkey.name, ngp_curline.name, NGP_MAX_NAME); /* and keyword's name */ ngp_linkey.name[NGP_MAX_NAME - 1] = 0; if (strlen(ngp_linkey.name) > FLEN_KEYWORD) /* WDP: 20-Jun-2002: mod to support HIERARCH */ { return(NGP_BAD_ARG); /* cfitsio does not allow names > 8 chars */ } return(NGP_OK); /* we have valid non empty line, so return success */ } } /* check whether keyword can be written as is */ int ngp_keyword_is_write(NGP_TOKEN *ngp_tok) { int i, j, r, l, spc; /* indexed variables not to write */ static char *nm[] = { "NAXIS", "TFORM", "TTYPE", NULL } ; /* non indexed variables not allowed to write */ static char *nmni[] = { "SIMPLE", "XTENSION", "BITPIX", "NAXIS", "PCOUNT", "GCOUNT", "TFIELDS", "THEAP", "EXTEND", "EXTVER", NULL } ; if (NULL == ngp_tok) return(NGP_NUL_PTR); r = NGP_OK; for (j = 0; ; j++) /* first check non indexed */ { if (NULL == nmni[j]) break; if (0 == strcmp(nmni[j], ngp_tok->name)) return(NGP_BAD_ARG); } for (j = 0; ; j++) /* now check indexed */ { if (NULL == nm[j]) return(NGP_OK); l = strlen(nm[j]); if ((l < 1) || (l > 5)) continue; if (0 == strncmp(nm[j], ngp_tok->name, l)) break; } if ((ngp_tok->name[l] < '1') || (ngp_tok->name[l] > '9')) return(NGP_OK); spc = 0; for (i = l + 1; i < 8; i++) { if (spc) { if (' ' != ngp_tok->name[i]) return(NGP_OK); } else { if ((ngp_tok->name[i] >= '0') || (ngp_tok->name[i] <= '9')) continue; if (' ' == ngp_tok->name[i]) { spc = 1; continue; } if (0 == ngp_tok->name[i]) break; return(NGP_OK); } } return(NGP_BAD_ARG); } /* write (almost) all keywords from given HDU to disk */ int ngp_keyword_all_write(NGP_HDU *ngph, fitsfile *ffp, int mode) { int i, r, ib; char buf[200]; long l; if (NULL == ngph) return(NGP_NUL_PTR); if (NULL == ffp) return(NGP_NUL_PTR); r = NGP_OK; for (i=0; itokcnt; i++) { r = ngp_keyword_is_write(&(ngph->tok[i])); if ((NGP_REALLY_ALL & mode) || (NGP_OK == r)) { switch (ngph->tok[i].type) { case NGP_TTYPE_BOOL: ib = ngph->tok[i].value.b; fits_write_key(ffp, TLOGICAL, ngph->tok[i].name, &ib, ngph->tok[i].comment, &r); break; case NGP_TTYPE_STRING: fits_write_key_longstr(ffp, ngph->tok[i].name, ngph->tok[i].value.s, ngph->tok[i].comment, &r); break; case NGP_TTYPE_INT: l = ngph->tok[i].value.i; /* bugfix - 22-Jan-99, BO - nonalignment of OSF/Alpha */ fits_write_key(ffp, TLONG, ngph->tok[i].name, &l, ngph->tok[i].comment, &r); break; case NGP_TTYPE_REAL: fits_write_key(ffp, TDOUBLE, ngph->tok[i].name, &(ngph->tok[i].value.d), ngph->tok[i].comment, &r); break; case NGP_TTYPE_COMPLEX: fits_write_key(ffp, TDBLCOMPLEX, ngph->tok[i].name, &(ngph->tok[i].value.c), ngph->tok[i].comment, &r); break; case NGP_TTYPE_NULL: fits_write_key_null(ffp, ngph->tok[i].name, ngph->tok[i].comment, &r); break; case NGP_TTYPE_RAW: if (0 == strcmp("HISTORY", ngph->tok[i].name)) { fits_write_history(ffp, ngph->tok[i].comment, &r); break; } if (0 == strcmp("COMMENT", ngph->tok[i].name)) { fits_write_comment(ffp, ngph->tok[i].comment, &r); break; } sprintf(buf, "%-8.8s%s", ngph->tok[i].name, ngph->tok[i].comment); fits_write_record(ffp, buf, &r); break; } } else if (NGP_BAD_ARG == r) /* enhancement 10 dec 2003, James Peachey: template comments replace defaults */ { r = NGP_OK; /* update comments of special keywords like TFORM */ if (ngph->tok[i].comment && *ngph->tok[i].comment) /* do not update with a blank comment */ { fits_modify_comment(ffp, ngph->tok[i].name, ngph->tok[i].comment, &r); } } else /* other problem, typically a blank token */ { r = NGP_OK; /* skip this token, but continue */ } if (r) return(r); } fits_set_hdustruc(ffp, &r); /* resync cfitsio */ return(r); } /* init HDU structure */ int ngp_hdu_init(NGP_HDU *ngph) { if (NULL == ngph) return(NGP_NUL_PTR); ngph->tok = NULL; ngph->tokcnt = 0; return(NGP_OK); } /* clear HDU structure */ int ngp_hdu_clear(NGP_HDU *ngph) { int i; if (NULL == ngph) return(NGP_NUL_PTR); for (i=0; itokcnt; i++) { if (NGP_TTYPE_STRING == ngph->tok[i].type) if (NULL != ngph->tok[i].value.s) { ngp_free(ngph->tok[i].value.s); ngph->tok[i].value.s = NULL; } } if (NULL != ngph->tok) ngp_free(ngph->tok); ngph->tok = NULL; ngph->tokcnt = 0; return(NGP_OK); } /* insert new token to HDU structure */ int ngp_hdu_insert_token(NGP_HDU *ngph, NGP_TOKEN *newtok) { NGP_TOKEN *tkp; if (NULL == ngph) return(NGP_NUL_PTR); if (NULL == newtok) return(NGP_NUL_PTR); if (0 == ngph->tokcnt) tkp = (NGP_TOKEN *)ngp_alloc((ngph->tokcnt + 1) * sizeof(NGP_TOKEN)); else tkp = (NGP_TOKEN *)ngp_realloc(ngph->tok, (ngph->tokcnt + 1) * sizeof(NGP_TOKEN)); if (NULL == tkp) return(NGP_NO_MEMORY); ngph->tok = tkp; ngph->tok[ngph->tokcnt] = *newtok; if (NGP_TTYPE_STRING == newtok->type) { if (NULL != newtok->value.s) { ngph->tok[ngph->tokcnt].value.s = (char *)ngp_alloc(1 + strlen(newtok->value.s)); if (NULL == ngph->tok[ngph->tokcnt].value.s) return(NGP_NO_MEMORY); strcpy(ngph->tok[ngph->tokcnt].value.s, newtok->value.s); } } ngph->tokcnt++; return(NGP_OK); } int ngp_append_columns(fitsfile *ff, NGP_HDU *ngph, int aftercol) { int r, i, j, exitflg, ngph_i; char *my_tform, *my_ttype; char ngph_ctmp; if (NULL == ff) return(NGP_NUL_PTR); if (NULL == ngph) return(NGP_NUL_PTR); if (0 == ngph->tokcnt) return(NGP_OK); /* nothing to do ! */ r = NGP_OK; exitflg = 0; for (j=aftercol; jtok[i].name, "TFORM%d%c", &ngph_i, &ngph_ctmp)) { if ((NGP_TTYPE_STRING == ngph->tok[i].type) && (ngph_i == (j + 1))) { my_tform = ngph->tok[i].value.s; } } else if (1 == sscanf(ngph->tok[i].name, "TTYPE%d%c", &ngph_i, &ngph_ctmp)) { if ((NGP_TTYPE_STRING == ngph->tok[i].type) && (ngph_i == (j + 1))) { my_ttype = ngph->tok[i].value.s; } } if ((NULL != my_tform) && (my_ttype[0])) break; if (i < (ngph->tokcnt - 1)) continue; exitflg = 1; break; } if ((NGP_OK == r) && (NULL != my_tform)) fits_insert_col(ff, j + 1, my_ttype, my_tform, &r); if ((NGP_OK != r) || exitflg) break; } return(r); } /* read complete HDU */ int ngp_read_xtension(fitsfile *ff, int parent_hn, int simple_mode) { int r, exflg, l, my_hn, tmp0, incrementor_index, i, j; int ngph_dim, ngph_bitpix, ngph_node_type, my_version; char incrementor_name[NGP_MAX_STRING], ngph_ctmp; char *ngph_extname = 0; long ngph_size[NGP_MAX_ARRAY_DIM]; NGP_HDU ngph; long lv; incrementor_name[0] = 0; /* signal no keyword+'#' found yet */ incrementor_index = 0; if (NGP_OK != (r = ngp_hdu_init(&ngph))) return(r); if (NGP_OK != (r = ngp_read_line(0))) return(r); /* EOF always means error here */ switch (NGP_XTENSION_SIMPLE & simple_mode) { case 0: if (NGP_TOKEN_XTENSION != ngp_keyidx) return(NGP_TOKEN_NOT_EXPECT); break; default: if (NGP_TOKEN_SIMPLE != ngp_keyidx) return(NGP_TOKEN_NOT_EXPECT); break; } if (NGP_OK != (r = ngp_hdu_insert_token(&ngph, &ngp_linkey))) return(r); for (;;) { if (NGP_OK != (r = ngp_read_line(0))) return(r); /* EOF always means error here */ exflg = 0; switch (ngp_keyidx) { case NGP_TOKEN_SIMPLE: r = NGP_TOKEN_NOT_EXPECT; break; case NGP_TOKEN_END: case NGP_TOKEN_XTENSION: case NGP_TOKEN_GROUP: r = ngp_unread_line(); /* WARNING - not break here .... */ case NGP_TOKEN_EOF: exflg = 1; break; default: l = strlen(ngp_linkey.name); if ((l >= 2) && (l <= 6)) { if ('#' == ngp_linkey.name[l - 1]) { if (0 == incrementor_name[0]) { memcpy(incrementor_name, ngp_linkey.name, l - 1); incrementor_name[l - 1] = 0; } if (((l - 1) == (int)strlen(incrementor_name)) && (0 == memcmp(incrementor_name, ngp_linkey.name, l - 1))) { incrementor_index++; } sprintf(ngp_linkey.name + l - 1, "%d", incrementor_index); } } r = ngp_hdu_insert_token(&ngph, &ngp_linkey); break; } if ((NGP_OK != r) || exflg) break; } if (NGP_OK == r) { /* we should scan keywords, and calculate HDU's */ /* structure ourselves .... */ ngph_node_type = NGP_NODE_INVALID; /* init variables */ ngph_bitpix = 0; ngph_extname = NULL; for (i=0; i=1) && (j <= NGP_MAX_ARRAY_DIM)) { ngph_size[j - 1] = ngph.tok[i].value.i; } } } switch (ngph_node_type) { case NGP_NODE_IMAGE: if (NGP_XTENSION_FIRST == ((NGP_XTENSION_FIRST | NGP_XTENSION_SIMPLE) & simple_mode)) { /* if caller signals that this is 1st HDU in file */ /* and it is IMAGE defined with XTENSION, then we */ /* need create dummy Primary HDU */ fits_create_img(ff, 16, 0, NULL, &r); } /* create image */ fits_create_img(ff, ngph_bitpix, ngph_dim, ngph_size, &r); /* update keywords */ if (NGP_OK == r) r = ngp_keyword_all_write(&ngph, ff, NGP_NON_SYSTEM_ONLY); break; case NGP_NODE_ATABLE: case NGP_NODE_BTABLE: /* create table, 0 rows and 0 columns for the moment */ fits_create_tbl(ff, ((NGP_NODE_ATABLE == ngph_node_type) ? ASCII_TBL : BINARY_TBL), 0, 0, NULL, NULL, NULL, NULL, &r); if (NGP_OK != r) break; /* add columns ... */ r = ngp_append_columns(ff, &ngph, 0); if (NGP_OK != r) break; /* add remaining keywords */ r = ngp_keyword_all_write(&ngph, ff, NGP_NON_SYSTEM_ONLY); if (NGP_OK != r) break; /* if requested add rows */ if (ngph_size[1] > 0) fits_insert_rows(ff, 0, ngph_size[1], &r); break; default: r = NGP_BAD_ARG; break; } } if ((NGP_OK == r) && (NULL != ngph_extname)) { r = ngp_get_extver(ngph_extname, &my_version); /* write correct ext version number */ lv = my_version; /* bugfix - 22-Jan-99, BO - nonalignment of OSF/Alpha */ fits_write_key(ff, TLONG, "EXTVER", &lv, "auto assigned by template parser", &r); } if (NGP_OK == r) { if (parent_hn > 0) { fits_get_hdu_num(ff, &my_hn); fits_movabs_hdu(ff, parent_hn, &tmp0, &r); /* link us to parent */ fits_add_group_member(ff, NULL, my_hn, &r); fits_movabs_hdu(ff, my_hn, &tmp0, &r); if (NGP_OK != r) return(r); } } if (NGP_OK != r) /* in case of error - delete hdu */ { tmp0 = 0; fits_delete_hdu(ff, NULL, &tmp0); } ngp_hdu_clear(&ngph); return(r); } /* read complete GROUP */ int ngp_read_group(fitsfile *ff, char *grpname, int parent_hn) { int r, exitflg, l, my_hn, tmp0, incrementor_index; char grnm[NGP_MAX_STRING]; /* keyword holding group name */ char incrementor_name[NGP_MAX_STRING]; NGP_HDU ngph; incrementor_name[0] = 0; /* signal no keyword+'#' found yet */ incrementor_index = 6; /* first 6 cols are used by group */ ngp_grplevel++; if (NGP_OK != (r = ngp_hdu_init(&ngph))) return(r); r = NGP_OK; if (NGP_OK != (r = fits_create_group(ff, grpname, GT_ID_ALL_URI, &r))) return(r); fits_get_hdu_num(ff, &my_hn); if (parent_hn > 0) { fits_movabs_hdu(ff, parent_hn, &tmp0, &r); /* link us to parent */ fits_add_group_member(ff, NULL, my_hn, &r); fits_movabs_hdu(ff, my_hn, &tmp0, &r); if (NGP_OK != r) return(r); } for (exitflg = 0; 0 == exitflg;) { if (NGP_OK != (r = ngp_read_line(0))) break; /* EOF always means error here */ switch (ngp_keyidx) { case NGP_TOKEN_SIMPLE: case NGP_TOKEN_EOF: r = NGP_TOKEN_NOT_EXPECT; break; case NGP_TOKEN_END: ngp_grplevel--; exitflg = 1; break; case NGP_TOKEN_GROUP: if (NGP_TTYPE_STRING == ngp_linkey.type) { strncpy(grnm, ngp_linkey.value.s, NGP_MAX_STRING); } else { sprintf(grnm, "DEFAULT_GROUP_%d", master_grp_idx++); } grnm[NGP_MAX_STRING - 1] = 0; r = ngp_read_group(ff, grnm, my_hn); break; /* we can have many subsequent GROUP defs */ case NGP_TOKEN_XTENSION: r = ngp_unread_line(); if (NGP_OK != r) break; r = ngp_read_xtension(ff, my_hn, 0); break; /* we can have many subsequent HDU defs */ default: l = strlen(ngp_linkey.name); if ((l >= 2) && (l <= 6)) { if ('#' == ngp_linkey.name[l - 1]) { if (0 == incrementor_name[0]) { memcpy(incrementor_name, ngp_linkey.name, l - 1); incrementor_name[l - 1] = 0; } if (((l - 1) == (int)strlen(incrementor_name)) && (0 == memcmp(incrementor_name, ngp_linkey.name, l - 1))) { incrementor_index++; } sprintf(ngp_linkey.name + l - 1, "%d", incrementor_index); } } r = ngp_hdu_insert_token(&ngph, &ngp_linkey); break; /* here we can add keyword */ } if (NGP_OK != r) break; } fits_movabs_hdu(ff, my_hn, &tmp0, &r); /* back to our HDU */ if (NGP_OK == r) /* create additional columns, if requested */ r = ngp_append_columns(ff, &ngph, 6); if (NGP_OK == r) /* and write keywords */ r = ngp_keyword_all_write(&ngph, ff, NGP_NON_SYSTEM_ONLY); if (NGP_OK != r) /* delete group in case of error */ { tmp0 = 0; fits_remove_group(ff, OPT_RM_GPT, &tmp0); } ngp_hdu_clear(&ngph); /* we are done with this HDU, so delete it */ return(r); } /* top level API functions */ /* read whole template. ff should point to the opened empty fits file. */ int fits_execute_template(fitsfile *ff, char *ngp_template, int *status) { int r, exit_flg, first_extension, i, my_hn, tmp0, keys_exist, more_keys, used_ver; char grnm[NGP_MAX_STRING], used_name[NGP_MAX_STRING]; long luv; if (NULL == status) return(NGP_NUL_PTR); if (NGP_OK != *status) return(*status); if ((NULL == ff) || (NULL == ngp_template)) { *status = NGP_NUL_PTR; return(*status); } ngp_inclevel = 0; /* initialize things, not all should be zero */ ngp_grplevel = 0; master_grp_idx = 1; exit_flg = 0; ngp_master_dir[0] = 0; /* this should be before 1st call to ngp_include_file */ first_extension = 1; /* we need to create PHDU */ if (NGP_OK != (r = ngp_delete_extver_tab())) { *status = r; return(r); } fits_get_hdu_num(ff, &my_hn); /* our HDU position */ if (my_hn <= 1) /* check whether we really need to create PHDU */ { fits_movabs_hdu(ff, 1, &tmp0, status); fits_get_hdrspace(ff, &keys_exist, &more_keys, status); fits_movabs_hdu(ff, my_hn, &tmp0, status); if (NGP_OK != *status) return(*status); /* error here means file is corrupted */ if (keys_exist > 0) first_extension = 0; /* if keywords exist assume PHDU already exist */ } else { first_extension = 0; /* PHDU (followed by 1+ extensions) exist */ for (i = 2; i<= my_hn; i++) { *status = NGP_OK; fits_movabs_hdu(ff, 1, &tmp0, status); if (NGP_OK != *status) break; fits_read_key(ff, TSTRING, "EXTNAME", used_name, NULL, status); if (NGP_OK != *status) continue; fits_read_key(ff, TLONG, "EXTVER", &luv, NULL, status); used_ver = luv; /* bugfix - 22-Jan-99, BO - nonalignment of OSF/Alpha */ if (VALUE_UNDEFINED == *status) { used_ver = 1; *status = NGP_OK; } if (NGP_OK == *status) *status = ngp_set_extver(used_name, used_ver); } fits_movabs_hdu(ff, my_hn, &tmp0, status); } if (NGP_OK != *status) return(*status); if (NGP_OK != (*status = ngp_include_file(ngp_template))) return(*status); for (i = strlen(ngp_template) - 1; i >= 0; i--) /* strlen is > 0, otherwise fopen failed */ { #ifdef MSDOS if ('\\' == ngp_template[i]) break; #else if ('/' == ngp_template[i]) break; #endif } i++; if (i > (NGP_MAX_FNAME - 1)) i = NGP_MAX_FNAME - 1; if (i > 0) { memcpy(ngp_master_dir, ngp_template, i); ngp_master_dir[i] = 0; } for (;;) { if (NGP_OK != (r = ngp_read_line(1))) break; /* EOF always means error here */ switch (ngp_keyidx) { case NGP_TOKEN_SIMPLE: if (0 == first_extension) /* simple only allowed in first HDU */ { r = NGP_TOKEN_NOT_EXPECT; break; } if (NGP_OK != (r = ngp_unread_line())) break; r = ngp_read_xtension(ff, 0, NGP_XTENSION_SIMPLE | NGP_XTENSION_FIRST); first_extension = 0; break; case NGP_TOKEN_XTENSION: if (NGP_OK != (r = ngp_unread_line())) break; r = ngp_read_xtension(ff, 0, (first_extension ? NGP_XTENSION_FIRST : 0)); first_extension = 0; break; case NGP_TOKEN_GROUP: if (NGP_TTYPE_STRING == ngp_linkey.type) { strncpy(grnm, ngp_linkey.value.s, NGP_MAX_STRING); } else { sprintf(grnm, "DEFAULT_GROUP_%d", master_grp_idx++); } grnm[NGP_MAX_STRING - 1] = 0; r = ngp_read_group(ff, grnm, 0); first_extension = 0; break; case NGP_TOKEN_EOF: exit_flg = 1; break; default: r = NGP_TOKEN_NOT_EXPECT; break; } if (exit_flg || (NGP_OK != r)) break; } /* all top level HDUs up to faulty one are left intact in case of i/o error. It is up to the caller to call fits_close_file or fits_delete_file when this function returns error. */ ngp_free_line(); /* deallocate last line (if any) */ ngp_free_prevline(); /* deallocate cached line (if any) */ ngp_delete_extver_tab(); /* delete extver table (if present), error ignored */ *status = r; return(r); }