From tsilveri Sun Apr 24 11:49:05 1994 Received: by cs.uml.edu id AA17269 (5.67a+/IDA-1.5); Fri, 22 Apr 1994 19:34:32 -0400 From: Bob Lechner Message-Id: <199404222334.AA17269@cs.uml.edu> Subject: Notes on genv7 bug meeting today 4/22 To: jhyun@cs.uml.edu, mcook@cs.uml.edu, mlamson@cs.uml.edu, hsano@cs.uml.edu, jmin@cs.uml.edu, mkeswani@cs.uml.edu, tsilveri@cs.uml.edu, schetty@cs.uml.edu, sharpana@cs.uml.edu, bguddeti@cs.uml.edu, fakhursh@cs.uml.edu, cyeng@cs.uml.edu, czhou@cs.uml.edu, wwenyen@cs.uml.edu, dolu@cs.uml.edu, namin@cs.uml.edu, dgrant@cs.uml.edu, tseeley@cs.uml.edu, wward@cs.uml.edu, jrichard@cs.uml.edu, lking@cs.uml.edu, mtorpey@cs.uml.edu, nosulliv@cs.uml.edu, zbrundag@cs.uml.edu, schiou@cs.uml.edu, dtong@cs.uml.edu, sislam@cs.uml.edu Date: Fri, 22 Apr 1994 19:34:31 -0400 (EDT) X-Mailer: ELM [version 2.4 PL22] Content-Type: text Content-Length: 4425 Status: OR CYeng and WWenyen met with me today regarding a genv7 test whose results we don't understand. Genv7 is used by OLCArch. The session revealed some problems with chgen v7 and potentially with gen v8. Genv7 has the 32-bit internal key format (type hcg_key) but not the pr_create/set/get/add sequence of calls to encapsulate access to table fields; the latter was introduced in chgen v6.5 and merged into genv8. Bde ignores version numbers and may not be affected. Meets_view location used/checked: tern(129)> grep meets_view *.c pr_load.c:112:int meets_view(tbl_idx,view_idx,pkey) tern(130)> pwd /nfs/cs/proj3/case/94s523/94solc/cyeng/build3 1. The mysterious bug: ---------------------- CYeng tried pr_adding a literal string constant instead of a string buffer variable, to resolve a problem with the OPclass 'add viewname table-row-string' operator command. This command failed to add a record with pkey BA010003 after pr_loading a database which already contained BA010001 and BA010002. More precise tests showed that BA010003 was not added when pr_add is passed the string entered by the operator, The reason is still a mystery. It may hve to do with the fact that in this case the usual call to pr_gen_pkey is bypassed. Any subsequent pr_gen_pkey(view, ttabbrev, ttkey) command will probably re-use BA010003 and overwrite the operator-added record. THus, the OPerator 'add' command needs fixing to 'do the right thing' by informing pr_gen_pkey to update its statistics on max pkey value for table BA. A generic fix to chgen requires more study. Similarly, any add commands contained in the script file need to maintain the integrity of the max pkey value in some way for pr_gen_pkey. Although the above fix is necessary in the long run, it will not necessarily fix the 'add' problem. We can't figure out why BA010003 is not added. NOTE: RUnning chgen -DDEBUG will cause pr_add to log to stdout every record that it adds to the database. Unfortunately the #ifdef DEBUG code merely prints the buffer argument to pr_add, not the actual contents of the struct that was added, if any. ------------------------------------------------------------------------ 2. The meets_view problem: ----------------------- In the course of inspecting pr_add and pr_dump we also discovered a problem with meets_view: Meets_view is now trivial: it returns the constant 1. It should be used by pr_add to reject any record whose type is not in the viewname passed as its first argument, or whose version number does not agree with the view's specification for that table. I suspect that meets_view was trivialized to avoid a problem with adding pkeys using version 0: this is not compatible with the logic of (former) meets_view because there was no way to select an absolute version number of 0. The chgen user guide specifies that the view contain a list of pairs, with the meanings: k = 0 meaning 'highest existing version#', whatever that is, k > 0 meaning 'absolute version# k', k < 0 meaning '(highest version#) - k' (k-th highest version) Meets_view was designed to reject any record that is not on the table list or that is on the list but has the wrong version number. This means that at most one version of a table can be pr_dumped per view. To dump two versions requires two pr_dump cals with distinct version numbers in the viewname arguments. The trivialized meets_view had no reason to be called, so there are no view table and version checks in chgem v7's utilities. This is not important now but will eventually have to be fixed (when view and version control are needed in chgen applications). [DEBUGGING NOTE: a view could be specifed containing tables EI and AI only, for use as a debug aid: e.g., dump the EI and AI tables at start and end of the k-th iteration of the main loop to files checkpoint-odd.k and checkpoint-even.k: if table EI contain any duplicates, the que is not completely consumed, deleted and re-filled during the cycle. The AI tables should identify each AI whose current state has been changed by actions during the cycle and (indirectly) each transition that was made and the event type that caused it.] ------------------------------------------------------------------------ 3. TBD: search genv6/7/8 reports for mention of bugs/fixes, and also for the reasons why genv7 trivialized and then dis-used function meets_view. From lechner Fri Apr 22 20:07:05 1994 Subject: More info onmeets_view To: lechner (Bob Lechner) Date: Fri, 22 Apr 1994 20:07:05 -0400 (EDT) Cc: cyyeng, wwenyen (Wenyen Wang(William)) X-Mailer: ELM [version 2.4 PL22] Content-Type: text Content-Length: 4535 Status: OR meets_view is discussed in final_report.tex in /nfs/cs/proj3/case/gen/ver_7/doc. There it is non-trivial (grep meets_view fin*.tex ==> line 439 and 2527): --------------------------------------------------------------------- Line 439: table_loop: calls: find_view_idx, meets_view Expected Result: loops through given table, stopping at all rows that meet the version for this view Correct Result: Yes meets_view: Expected Result: Status indicating whether given row is visible in a given view Correct Result: Yes ------------------------------------------------------------------ Line 2526: \begin{verbatim} int meets_view(tbl_idx, view_idx, pkey) \end{verbatim} This function is always called with a form of tbl\#\#curr->tbl\#\#id Thus, it is always passed an integer key. \vspace{1ex} \noindent Old code: \begin{verbatim} int tbl_idx, view_idx; char *pkey; { char ver_temp[3]; strncpy(ver_temp, pkey+2,2); ver_temp[2] = '\0'; return(hcg_view_list.view_list[view_idx].version_list[tbl_idx] == (char) atoi(ver_temp)); } \end{verbatim} \noindent New code: \begin{verbatim} int tbl_idx, view_idx; key_t pkey; { return(hcg_view_list.view_list[view_idx].version_list[tbl_idx] == (get_version(pkey))); } \end{verbatim} ----------------------------------------------------------------------- Here is what the source code contains: tern(171)> pwd /nfs/cs/proj3/case/gen/ver_7/src/SCCS tern(172)> grep meets_view s.*.c s.gen_defines.c:1408: fprintf(schh_fp," if ((meets_view(tbl%s_idx,hcg_view_idx,tbl%scurr->tbl%sid)) &&\\\n",sep,sep,sep); s.gen_defines.c:1427: fprintf(schh_fp," if ((meets_view(tbl%s_idx,hcg_view_idx,tbl%scurr->tbl%sid)) &&\\\n",sep,sep,sep); s.gen_defines.c:1446: fprintf(schh_fp," if ((meets_view(tbl%s_idx,hcg_view_idx,tbl%scurr->tbl%sid)) &&\\\n",sep,sep,sep); s.gen_defines.c:1471: fprintf(schh_fp," if (meets_view(tbl%s_idx,hcg_view_idx,tbl%scurr->tbl%sid))\n\n",sep,sep,sep); s.gen_defines.c:1514: fprintf(schh_fp," if ((meets_view(tbl%s_idx,hcg_view_idx,tblvar->tbl%sid)) &&\\\n",sep,sep); s.gen_defines.c:1532: fprintf(schh_fp," if ((meets_view(tbl%s_idx,hcg_view_idx,tblvar->tbl%sid)) &&\\\n",sep,sep); s.gen_defines.c:1557: fprintf(schh_fp," if (meets_view(tbl%s_idx,hcg_view_idx,tblvar->tbl%sid))\n\n",sep,sep); Of particular interest is this line: s.gen_pr_utils.c:20:c Added additional fix to meets_view for zero version problem s.gen_pr_utils.c:242: fprintf(prload_fp, "int meets_view(tbl_idx,view_idx,pkey)\n"); tern(173)> ------------------------------------------------------------------------ Line 20 says: ^As 00004/00000/00248 ^Ad D 1.4 92/12/22 10:46:27 cmurphy 4 3 ^Ac Added additional fix to meets_view for zero version problem -------------------------------------------------------------------- gen_pr_utils (via sccs get from the above SCCS directory) contains the nontrivial definition of meets_view (latest version): fprintf(prload_fp, "/******************************************************************************/\n"); fprintf(prload_fp, "/* This routine will extract a version number and row_number out of the */\n"); fprintf(prload_fp, "/* specified key value. The results are stored in the globals hcg_version */\n"); fprintf(prload_fp, "/* and hcg_row. */\n"); fprintf(prload_fp, "/******************************************************************************/\n"); fprintf(prload_fp, "/* This function will determine if the specified key value meets the */\n"); fprintf(prload_fp, "/* proper version for that table in the specified view idx. */\n"); fprintf(prload_fp, "/******************************************************************************/\n"); /**** 93s523 ****/ fprintf(prload_fp, "int meets_view(tbl_idx,view_idx,pkey)\n"); fprintf(prload_fp, "int tbl_idx;\n"); fprintf(prload_fp, "int view_idx;\n"); fprintf(prload_fp, "key_t pkey;\n"); fprintf(prload_fp, "{\n"); #ifndef NEW_VERSION fprintf(prload_fp," return(hcg_view_list.view_list[view_idx].version_list[tbl_idx] == (get_version(&pkey)))\n"); #else fprintf(prload_fp," return(1);\n"); #endif fprintf(prload_fp, "}\n\n"); ---------------------------------------------------------------- Therefore it is still a mystery why meets_view was trivialized. From lechner Sat Apr 23 00:08:07 1994 Received: by cs.uml.edu id AA01821 (5.67a+/IDA-1.5); Sat, 23 Apr 1994 00:08:06 -0400 From: Bob Lechner Message-Id: <199404230408.AA01821@cs.uml.edu> Subject: NEW_VERSION flag To: cyeng (Cheng-Lu Louis Yeng), wwenyen (Wenyen Wang(William)) Date: Sat, 23 Apr 1994 00:08:05 -0400 (EDT) Cc: lechner (Bob Lechner) X-Mailer: ELM [version 2.4 PL22] Content-Type: text Content-Length: 1142 Status: OR #Ifdef NEW_VERSION controls meets_view definition. THe makefile in /nfs/cs/proj3/case/gen/ver_7/test/schooldb8 omits it: --------------------------------------------------- tern(557)> cat Makefile # This Makefile was generated by maker: the makefile generator (c) 1991 AJS. # This Makefile is in the public domain -AJS # There is NO WARRANTY on this software... # CFLAGS = -DNEW_VERSION # LDFLAGS = -s # Comment these lines out to avoid info for the debugger. CC = /bin/cc RM = /bin/rm -f OFILES = schooldb.o pr_delete.o pr_load.o pr_dump.o pr_stats.o pr_free.o LIBS = EXEC-NAME = schooldb all: ${EXEC-NAME} ${EXEC-NAME}: ${OFILES} ${CC} ${LDFLAGS} -o $@ ${OFILES} ${LIBS} clean: ${RM} ${OFILES} ${EXEC-NAME} core a.out tern(558)> pwd /nfs/cs/proj3/case/gen/ver_7/test/schooldb8 tern(559)> -------------------------------------------------- Unfortunately the pr_*.c code in those testcase/src directories does NOT reference meets_view: it is onluy declared: tern(569)> grep meets_v pr*.c pr_load.c:112:int meets_view(tbl_idx,view_idx,pkey) tern(570)> pwd /nfs/cs/proj3/case/gen/ver_7/test/schooldb8 tern(571)> --