Many hyperlinks are disabled.
Use anonymous login
to enable hyperlinks.
16 check-ins using file embedded/www/doc/files/marpatcl_changes.html version 81f9b2f3fd
|
2018-04-19
| ||
| 19:47 | Update main line with the accumulated memory fixes. check-in: dcf340cd32 user: aku tags: trunk | |
| 19:44 | Merged series of fixes for memory issues (memory leaks, management mismatches, ...). Together with the push of memory-intensive tests into child processes done here the testsuite should now be properly constrained again wrt memory usage. Updated tests. Closed-Leaf check-in: dc34e1afbb user: aku tags: mem-limit | |
| 18:18 | Added test against unbound parser memory usage in parser-core. Found and fixed SV ref-count mismanagement (RCM) in the lex-only code path of the RTC. Added narrative tracing to the code path. Cleanup of lexer-core testsuite with regard to memory debugging. Found and fixed Tcl_Obj* RCM in the lex-only critcl template. Added narrative tracing to the template. Closed-Leaf check-in: a29c4613b0 user: aku tags: memory-cleanup | |
|
2018-04-18
| ||
| 23:41 | Do not mix malloc and Tcl allocation routines. Even when the malloc is hidden, here in `strdup`. Used to allocate strings with strdup, then release with ckfree/Tcl_Free. This messed up memory mgmt internals to the point of memory smashes. Fixed, now using our own implementation of strdup based on the allocation macros from environment.h. This ensures that alloc and free match. Thank you, valgrind. check-in: 714d438603 user: aku tags: memory-cleanup | |
| 03:08 | Moved the main parts of `test/support/gen.tcl` into `bin/i-gen`. This new internal generator application uses the public `marpa-gen` as the underlying workhorse. The support code now invokes the internal generator instead of doing everything itself. With the operation confined to a child process the amount of memory taken by the test process' is limited to the generated package itself instead of all the packages needed to perform the generation. Further modified lexer-core to place the sets of tests for each variant grammar/lexer-action into a child process as well. This ensures that the memory needed by each variant grammar is limited to that child process, instead of accumulating in the controlling test process. __Attention__: This change requires a Kettle with support for `kt::sub`, added to Kettle with commit [ef384673c5] (2018-04-18 02:28:17). check-in: ddc1e67640 user: aku tags: mem-limit | |
|
2018-04-17
| ||
| 20:36 | Moved the main parts of `test/support/gen.tcl` into `bin/i-gen`. This new internal generator application uses the public `marpa-gen` as the underlying workhorse. The support code now invokes the internal generator instead of doing everything itself. With the operation confined to a child process the test process' amount of memory taken is limited to the generated package instead of also keeping all the overhead of generating Note, this does not help with the test suites based on lexer-core as that suite still loads/sources ten different lexer packages into the same process. Handling this requires more work, i.e. pushing the individual tests into their own child processes. check-in: ad4f1d4287 user: aku tags: mem-limit | |
|
2018-04-11
| ||
| 23:33 | Extended the narrative tracing in `sem_tcl.c` to track refcounts. Found and fixed the cause for the orphaned Tcl_Obj*'s. The function `marpatcl_rtc_sv_astcl` did an extraneous refcount increment on the conversion result (SV tree to Tcl_Obj* tree). Parser operation based on RTC now does not leak anything anymore. check-in: 302f47227a user: aku tags: memory-cleanup | |
| 07:43 | Added code (sem_debug.c) to track SV allocation and release, and dump orphan SVs at the end. All SVs were orphaned, nothing released. Tracked to a bad guard condition in function `marpatcl_rtc_sv_unref`. As the code checks the refcount before decrementing it both 0 and 1 must trigger destruction, not just 0. Fixed. All SV structures are now properly released at the end (parser instance destruction). Still orphaned things left, these however are Tcl_Obj's. check-in: 0d6e6c1dbe user: aku tags: memory-cleanup | |
|
2018-04-10
| ||
| 04:55 | container: Fixed leakage of priority and priority alternate objects. Test support extended with various diagnostic tools. Note, it looks like the RTC and glue into Tcl are also leaking like mad, especially around the semantic values and the (partial) ASTs we are constructing. Continue investigation and fixing. Started a new branch for this, check-in: c3ff9015b6 user: aku tags: memory-cleanup | |
|
2018-04-05
| ||
| 08:29 | runtime-tcl, gate: Moved a few code blocks into their own methods to make their now-caller more readable. check-in: 0dff9cc32b user: aku tags: trunk | |
|
2018-04-03
| ||
| 20:37 | Fixed bug in the semantics' handling of :lexeme. Do not exclude the symbol from LATM fixup if the :lexeme adverbs do _not specify_ latm information. Facepalm. Found playing with lexeme events, and GC missing the crucial latm information, breaking generated test parsers. Updated tests to correct results. check-in: e5e442db2a user: aku tags: trunk | |
| 17:47 | Tests: Remove a leftover break from debugging something, wrongly committed. check-in: 7c7353ad24 user: andreask tags: trunk | |
|
2018-03-30
| ||
| 23:38 | Get doc fixes check-in: d41168399d user: aku tags: trunk | |
| 23:37 | Docs: Typo fixes, phrasing fixes. check-in: 2a1acc9132 user: aku tags: docs | |
| 23:24 | Make recent documentation work official. check-in: 607761c40d user: aku tags: trunk | |
| 23:23 | Docs work - Moved architecture from intro to dev guide. - Made intro a dispatcher to other documents based on readers goals. - Added reference for marpagen. - Added placeholder for SLIF, referencing the upstream Marpa::R2 documentation. check-in: 9f10a242a4 user: aku tags: docs | |