Commit message (Collapse) | Author | Age | Files | Lines | |
---|---|---|---|---|---|
* | Many improvements to code generation of efficient break and continue. | dkf | 2013-06-02 | 4 | -81/+235 |
| | |||||
* | Getting better at doing more efficient break/continue instruction handling. | dkf | 2013-06-01 | 3 | -50/+111 |
| | |||||
* | derp | dkf | 2013-05-30 | 1 | -2/+2 |
| | |||||
* | Working towards the next batch of optimizations. | dkf | 2013-05-30 | 4 | -19/+168 |
| | |||||
* | Corrected code generation when doing the second run with an 'infinite' loop. | dkf | 2013-05-30 | 1 | -1/+1 |
| | |||||
* | Simplifications and tidying up of stack management issues. Work in progress. | dgp | 2013-05-29 | 1 | -14/+13 |
| | |||||
* | 3614102 - Reset stack housekeeping when compileProc fails. | dgp | 2013-05-29 | 1 | -2/+6 |
| | |||||
* | Stop emitting the instructions INST_*_SCALAR_STK*. They are identical to | dgp | 2013-05-29 | 6 | -94/+67 |
| | | | | | | | their INST_*_STK* counterparts. Having done that, it is clear the "simpleVarName" return from TclPushVarName provides nothing of use to any of its callers. Eliminate that. Also make TPVN return void, instead of int. Bring the TPVN header comments up to date; they were quite rotten. | ||||
* | Use the routines that provide "basic compile" instead of reinventing them. | dgp | 2013-05-28 | 1 | -34/+16 |
| | |||||
* | Plug the memory leak. Greater ambitions to improve this routine are proving | dgp | 2013-05-28 | 1 | -0/+1 |
| | | | more difficult than expected. | ||||
* | Missed bits of dup code elimination. | dgp | 2013-05-28 | 1 | -23/+0 |
| | |||||
* | Repair some stack depth housekeeping. | dgp | 2013-05-25 | 4 | -7/+3 |
| | |||||
* | 3613854 - Fixup stack maintenance /code generation for [array set x $oddList]. | dgp | 2013-05-24 | 1 | -20/+17 |
| | | | | Postscript - I see that this commit created a memory leak. Will commit a fix within a few days. | ||||
* | Spare developers the burden and error risk of counting bytes in string | dgp | 2013-05-23 | 4 | -77/+81 |
| | | | literals, or having to type them twice. | ||||
* | Eliminate code duplication. | dgp | 2013-05-23 | 4 | -666/+68 |
| | |||||
* | Removed const qualifier causing the HP native cc to error out (error 1675: ↵ | andreask | 2013-05-22 | 1 | -1/+1 |
| | | | | Duplicate type qualifier "const"). | ||||
* | silence compiler warning | dgp | 2013-05-22 | 1 | -1/+2 |
|\ | |||||
| * | silence compiler warning | dgp | 2013-05-22 | 1 | -1/+2 |
| | | |||||
* | | [3613609]: Replace strcasecmp() with UTF-8-aware version. | dkf | 2013-05-22 | 4 | -6/+47 |
|\ \ | |/ | |||||
| * | Fixed the weird edge case.bug_3613609 | dkf | 2013-05-22 | 1 | -12/+25 |
| | | |||||
| * | Slight improvement: if cs = "\xC0\x80" and ct = "\x00", loop would continue ↵ | jan.nijtmans | 2013-05-21 | 1 | -4/+4 |
| | | | | | | | | after NUL-byte, this should not happen. | ||||
| * | Proposed solution for 3613609: lsort -nocase does not sort non-ASCII correctly | jan.nijtmans | 2013-05-21 | 4 | -6/+34 |
| | | |||||
* | | 3613569 Handle case when TclpTempFileNameForLibrary returns NULL. | dgp | 2013-05-20 | 1 | -0/+3 |
| | | |||||
* | | Don't #define VOID on VxWorks, as it is already typdef'd to void. Eliminate ↵ | jan.nijtmans | 2013-05-19 | 4 | -19/+14 |
|\ \ | |/ | | | | | possibly conflicting LOCAL/OFFSET | ||||
| * | Don't #define VOID on VxWorks, as it is already typdef'd to void. Eliminate ↵ | jan.nijtmans | 2013-05-19 | 3 | -11/+13 |
| |\ | | | | | | | | | | possibly conflicting LOCAL/OFFSET | ||||
| | * | Don't #define VOID on VxWorks, as it is already typdef'd to void | jan.nijtmans | 2013-05-19 | 2 | -7/+9 |
| | | | | | | | | | Eliminate possibly conflicting LOCAL #define | ||||
| | * | <sys/stat.h> inclusion is only needed when compiling for Win32, don't bother ↵ | jan.nijtmans | 2013-05-17 | 1 | -1/+1 |
| | | | | | | | | | | | | for other platforms. | ||||
| | * | - eliminate compiler warning when compiling with Visual Studio. | jan.nijtmans | 2013-05-17 | 1 | -0/+1 |
| | | | | | | | | | - Make sure that _ftime() from msvcrt.dll is used, not ftime() from mingw (which might use 64-bit time_t) | ||||
* | | | Split tclCompCmds.c into two roughly-equal-sized pieces. | dkf | 2013-05-18 | 2 | -2931/+3276 |
| | | | |||||
* | | | Remove useless variables. | dkf | 2013-05-17 | 1 | -42/+39 |
|\ \ \ | |||||
| * | | | Confirmed that every caller of TclProcCompileProc() arranges for the | dgp | 2013-05-16 | 1 | -11/+2 |
| | | | | | | | | | | | | | | | | | | | | | | | | procPtr and nsPtr arguments: nsPtr == procPtr->cmdPtr->nsPtr. This makes the test in TclCompileScript() useless. TCS() will always compile in the current namespace of the interp. Remove the code that obfuscates that fact. | ||||
| * | | | Disabled some code in TclCompileScript(). Test suite results are unaffected. | dgp | 2013-05-15 | 1 | -1/+3 |
| | | | | | | | | | | | | Does this indicate a gap in the test suite, or is this code truly useless? | ||||
* | | | | Add support for the latest mingw-4.0-rc1. See: ↵ | jan.nijtmans | 2013-05-16 | 1 | -0/+11 |
|\ \ \ \ | |/ / / |/| / / | |/ / | [http://sourceforge.net/p/mingw/mingw-org-wsl/ci/4.0-rc1/tree/NEWS] | ||||
| * | | _USE_32BIT_TIME_T is already defined in tclWinPort.h | jan.nijtmans | 2013-05-16 | 1 | -4/+0 |
| | | | |||||
| * | | Add support for the latest mingw-4.0-rc1. See: ↵ | jan.nijtmans | 2013-05-16 | 1 | -0/+15 |
| |\ \ | | |/ | | | | | | | [http://sourceforge.net/p/mingw/mingw-org-wsl/ci/4.0-rc1/tree/NEWS] | ||||
| | * | Add support for the latest mingw-4.0-rc1. See: ↵ | jan.nijtmans | 2013-05-16 | 1 | -0/+15 |
| | | | | | | | | | | | | [http://sourceforge.net/p/mingw/mingw-org-wsl/ci/4.0-rc1/tree/NEWS] | ||||
* | | | Removing a few changes that were not actually needed, and correcting comments. | dkf | 2013-05-15 | 2 | -10/+7 |
| | | | |||||
* | | | merge trunkdkf_bcc_optimize | dkf | 2013-05-15 | 2 | -3/+4 |
|\ \ \ | |||||
| * \ \ | Merged fix to bytecode engine problem. Details in merged revision. | andreask | 2013-05-13 | 1 | -2/+3 |
| |\ \ \ | |||||
| | * | | | Fixed bug in parent revision [832a1994c7] unpredictably breaking thebugfix_832a1994c7_for_precompiled_bc | andreask | 2013-05-13 | 1 | -2/+3 |
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | execution of precompiled bytecode (i.e. tbcload'ed bytecode). The semi-inversion of a guard condition dropped the part about TCL_BYTECODE_PRECOMPILED of the original condition, allowing the core to attempt to recompile code for which there are no script sources (anymore), which then fails. | ||||
| * | | | | compiler warning | dgp | 2013-05-13 | 1 | -1/+1 |
| | | | | | |||||
* | | | | | A better technique for [list {*}blah]. Remove the INST_LIST_EXPANDED opcode ↵ | dkf | 2013-05-15 | 6 | -150/+198 |
|/ / / / | | | | | | | | | | | | | (and the complex machinery associated with it) as as it is no longer needed. | ||||
* | | | | Corrected the stack balancing in the special [list {*} ] compiler. | dkf | 2013-05-12 | 1 | -1/+9 |
| | | | | |||||
* | | | | Fix implementation of INST_LIST_EXPANDED. | dkf | 2013-05-12 | 1 | -1/+5 |
| | | | | |||||
* | | | | Partial fix: still ongoing | dkf | 2013-05-11 | 2 | -1/+4 |
| | | | | |||||
* | | | | Optimizations and general bytecode generation improvements. | dkf | 2013-05-10 | 5 | -53/+535 |
| | | | | |||||
* | | | | Add support for Cygwin64, which has a 64-bit "long" type. | jan.nijtmans | 2013-05-06 | 2 | -0/+133 |
|\ \ \ \ | | |/ / | |/| | | | | | | Binary compatibility with win64 requires that all stub entries use 32-bit long's, therefore the need for various wrapper functions/macros. For Tcl 9 a better solution is needed, but that cannot be done without introducing binary incompatibility. | ||||
| * | | | Add support for Cygwin64, which has a 64-bit "long" type. | jan.nijtmans | 2013-05-06 | 2 | -0/+133 |
| |\ \ \ | | | |/ | | |/| | | | | | Binary compatibility with win64 requires that all stub entries use 32-bit long's, therefore the need for various wrapper functions/macros. For Tcl 9 a better solution is needed, but that cannot be done without introducing binary incompatibility. | ||||
| | * | | Add support for Cygwin64, which has a 64-bit "long" type. | jan.nijtmans | 2013-05-06 | 2 | -0/+133 |
| | | | | | | | | | | | | | | | | Binary compatibility with win64 requires that all stub entries use 32-bit long's, therefore the need for various wrapper functions/macros. For Tcl 9 a better solution is needed, but that cannot be done without introducing binary incompatibility. | ||||
| * | | | Backport 8.6's fix [checkin 5af0d249de] to [Bug 2901998]: Inconsistent ↵ | ferrieux | 2013-05-01 | 1 | -8/+92 |
| | | | | | | | | | | | | | | | | buffered I/O. Tcl's I/O now flushes buffered output before reading, discards buffered input before writing, etc. |