Copyright (C) 2023-2024 Frank Eskesen.
This file is free content, distributed under the MIT license. (See the accompanying file LICENSE.MIT or the original contained within https://opensource.org/licenses/MIT)
Minor changes are not documented in this change log, but since the distribution is maintained in git, changes are always recorded.
Added auto-recompile trigger for dev and pub libraries to account for a pub::Object::~Object linkage change.
Restructured ~/src/cpp/Brian/.
- Updated the Command and Service objects.
- Added USE_STATIC environmental control to select whether to build a STATIC or a SHARED (dynamic) library.
Regardless of whether static or shared library linkage is chosen, simply adding an object module containing a static Command or Service object automatically adds that Command or Service to its map.
Removed boost::core::demangle.hpp dependency.
Merged maint/preview into trunk.
The combined editors (xcbedit and xtmedit) are now ready.
The Utf.h documentation is ready for github markdown verification.
The documentation updates and most of the code changes in this commit are interrelated. The Symbol type was particularly helpful in simplifying the documentation while also clarifying the interface definitions.
The combined editor is now in production test, with no known new errors.
We spent a lot of time trying to implement common REDO functions in EdRedo.hpp. This was wasted effort: callers still needed to test whether or not an insert after a line with a binary delimiter occurred and handle it differently in each case.
The include command (in EdBifs.cpp) was too difficult to fix. Rather than botching the file when including a file after a binary line, now it's simply disallowed. The line insert command (Alt-I) can be used to repair the null-delimited binary line, then a file insert after it works properly.
Object counter tests were added to detect memory leaks. So far, so good.
We'll use the updated editor to restructure it. Redo, line, and message functions shouldn't be in EdFile. While this restructure isn't strictly necessary, it will be a good production test. The Utf documentation has been in progress for a while. This and a full editor function test need to be completed before a trunk release.
The combined editor is now in production test, with no known new errors. Known error: include command after a line with a null terminator does not build the undo/redo object correctly.
Added fixes: src/cpp/Edit/Xcb, src/cpp/Edit/Term
Renamed ~/src/cpp/Edit/Xcb/editxcb to xcbedit and ~/src/cpp/Edit/Term/editerm to xtmedit. This also required changes to some scripts in ~/bat/.
Removed obsolete Utf objects. Moved ~/src/cpp/inc/pub/Utf.i to ~/src/cpp/inc/pub/bits/Utf.i. Moved ~/src/cpp/inc/pub/bits/Utf_types.i to ~/src/cpp/inc/pub/Utf.i. ~/pub/Utf.i is now the include used to define the Utf.h types in the application's namespace.
Removed utf*_decoder set_offset and added set_cpoint method. This removes the need for OFFSET_COLUMN, and is a cleaner and more logical interface.
The editor modules have been updated to use the updated decoders and encoders. These modules are now in beta test mode. A complete function test is still required.
Revised utf*_decoder and utf*_encoder interfaces so that the current column always matches the column as set by set_column(). This is a cleaner and more logical interface.
Since method decode() updates the offset and column number, buffer[offset] applies to the column number of the next codepoint. Method set_column() updates both the column and offset fields. With this change buffer[offset] is the first character of a column, not one past that character.
The editor modules have not been updated yet. They (still) do not use the updated decoders and encoders.
~/src/cpp/Edit/Xcb/* and ~/src/cpp/Edit/Term/* modules need to use the updated decoders and encoders.
This has no visible effect for Xcb/Editor modules but does require changes. This editor does not support Utf combining characters. It writes them as separate characters.
The (ncurses based) Term/Editor output already combined Utf combining characters, but did not properly position the cursor. Since the cursor wasn't accurate for combining characters, as a safety measure we partially disabled Utf support. Files containing non-ASCII Unicode characters were marked as damaged so they could not be overwritten by the editor.
Completely tested all portions of all the utf*_decoder and utf*_encoder implementations.
Fixed a minor library makefile glitch: Each library also had a build dependency on its Test subdirectory, causing unnessary make operations when that subdirectory changed.
Updated ~/src/cpp/lib/pub/Utf.cpp and ~/src/cpp/inc/pub/Utf.h. Added ~/src/cpp/lib/pub/Test/Test_utf.cpp and ~/src/cpp/inc/pub/bits/Utf.i.
The Utf.h updates use user allocated buffers for encoding and decoding rather than the current automatically allocated buffers. Test_utf.cpp partially tests the encoding and decoding portions of the newly implemented code.
~/src/cpp/Edit/Xcb/EdOuts.cpp tested the new implementation in production code. (The testing code was included but not activated in this commit.)
The ncurses and xcb editors are now nominally operational.
Too much is working not to share this version, but too much is not working well enough for this to be a trunk commit.
The xcb editor has been refactored so that almost all functionality is shared with the ncurses editor.
The EdInps and EdOuts modules are not shared. They provide the input and output function for each of the editors.
The ncurses editor is compiled with CDEFS += -DUSE_NCURSES_VERSION added
to the makefile.
This is only used in one place: In Edit.cpp function main(), the opt_bg
option is ignored in the ncurses version.
The curses editor always runs in the main thread.
What's not working:
- UTF-8 support.
When running an xterm a Fedora machine (locally or remotely,) a UTF-8 character sequence isn't recognized as UTF at all. The easiest way to explain the result is that it's not usable. You get jibberish on lines and the next line's display can be corrupted. This certainly has to be fixed.
On Cygwin, UTF-8 sequences are displayed correctly, even including combined character sequences. Unfortunately, code that we rely on to position the cursor doesn't recognize these sequences. Since our cursor is mis-positioned at every point on the line after a combined sequence, editing becomes problematic.
Neither the xcb editor nor the pub library handle combined characters, so at least they agree on character positioning and editing works. However, they both need to be fixed to handle these combined characters. This looks like a difficult job.
- Cursor support
The ncurses editor currently lets ncurses take care of cursor positioning, so the cursors look different. We'll probably need to turn off ncurses cursor positioning and just use the hide and show_cursor methods. This should be but hasn't been relatively easy.
- Mouse cursor hiding
The xcb editor appears to have regressed. Hiding no longer works. The ncurses editor doesn't have this function.
Library restructure stable and ready for trunk merge.
Updated editxcb, restructuring and adding function. See: .README
- The ~/obj/cpp/NN2 make process now writes an error message (but completes normally) if ImageMagick-7 is not installed. User modification of the makefile is not required.
This is a maintenance commit.
Existing known anomalies:
- Running make compile, ~/obj/cpp/Stock once attempted to link before the object files were compiled. (Not reproducible.)
- ~/obj/cpp/HTTP/SampleSSL loops. (Not diagnosed.)
Changes:
- The C++ library build has been restructured.
- Static libraries are built in ~/obj/cpp/lib/static
- Shared libraries are built in ~/obj/cpp/lib/shared
- In CYGWIN, the shared libraries are DLLs rather than unix libraries
- Documentation has been updated. This work is incomplete.
- ~/bat/.home contains recommended bash initialization scripts.
The C++ library build needs more testing, especially in regard to dependency resolution. Perhaps the ${module} variable can be used to make the Makefile.BSD more common.
- Partially updated Debug.h documentation, correcting links and missing information.
- Removed pub::Debug::clr_head and changed pub::Debug::set_head to set complete heading rather than set heading bit.
- Fixed ~/src/cpp/lib/dev/Test/T_Stream.hpp, adding a short delay to allow threads to complete. (Otherwise Stream object count error could occur.)
- Fixed ~/src/cpp/lib/dev/Test/script/regression.d/.prereq.sh and ~/src/cpp/lib/pub/Test/script/regression.d/.prereq.sh, making ~/obj/cpp/lib rather than just the associated library.
- Fixed static library build, removing include of dependent libraries. This was just wasted filler.
- Corrected some static library build dependencies.
- Fixed one minor problem in Makefile (Remove operations are not displayed.)
The C++ library build has been modified so that the shared and static libraries use the same set of object files. The static libraries are copied to ~/obj/cpp/lib/static/. The shared libraries are copied to ~/obj/cpp/lib/shared/.
When building, your L link that points to the library should point to the static or shared library.
To use the C++ shared library: Your "L" library link should link to ~/.local/lib/sdlc++ (which is a link to ~/obj/cpp/lib/shared created during shared library installation.)
On Cygwin, add ~/.local/lib/sdlc++ to your PATH On Linux, add ~/.local/lib/sdlc++ to your LD_LIBRARY_PATH
To use the C++ static library: Your "L" (library link) should link to either ~/obj/cpp/L, ~/obj/cpp/lib/L, or ~/obj/cpp/lib/static/. It SHOULD NOT link to ~/obj/cpp/lib, which contains both the static and (incomplete) shared library objects.
Note: This is a test commit.
- Updated makefiles
- C++ libraries now use the same object files, building static libraries in ~/obj/cpp/lib/static and shared libraries in ~/obj/cpp/lib/shared/.
- Use "@rm -f ..." instead of "-@rm ... >/dev/null 2>/dev/null" (Even updating the now unsupported Windows makefiles.)
- Empty last makefile lines have been removed.
- The python library is now built using standard python mechanisms rather than creating a link to the library source.
- Added "make update" feature.
- File ~/bat/sys/_want.version contains a list of desired versions
- Various libraries and components contain _have.version version files
- We update the _want.version file when a recompile is needed even when include dependencies are met.
- Make update recompiles the updated components.
- Doxygen now built from ~/doc/obj/Doxygen
- This is the only documentation for the COM, DEV, GUI, and OBJ libraries
- Only the PUB library has any Doxygen markup, and that is minimal. Githup markup documentation in progress for the PUB library.
- Documentation update test (Verify .md files on github.)
- Removed Debug::errorp method. (Moved to ~/src/cpp/lib/pub/Fileman.cpp)
- XCB editor (~/src/cpp/Edit/Xcb)
- New feature: ctrl-F9 copies current line to history line. (F9 without ctrl copies the filename to the history line.)
- src/cpp/RFC: RFC7541.*
- Almost all functions fully operational. (Pack::resize does not yet transfer the resize request via an Ioda Writer/Reader.)
- Added Ioda::dump function. It displays the entire Ioda buffer sequence.
- XCB editor (~/src/cpp/Edit/Xcb)
- New feature: If not an editor command, run as a system command and display output in a new pseudo file.
- New features: 'help' command and 'set help' command.
- New features: Added common ctrl- keys: ctrl-S, ctrl-Q, ctrl-Y, and ctrl-Z.
- Added 'TIO' misspelling of 'TOP' command.
- List.h (~/src/cpp/inc/pub/List.h, ~/src/cpp/inc/pub/bits/List.h
- Fixed SHSL_list iterator. It shouldn't be destructive.
- Added '= delete' constructors and assignment operators. Since all lists use link references rather than copies, none of the list objects can be copied. List objects can be moved, but that functionality isn't needed (yet.)
- src/cpp/RFC: RFC7541.*
- Huffman decoding and encoding are fully operational. (Timing tests and production tests have not been run.)
- HPACK compression has a preliminary interface defined. The interface implementation is scaffolded and compile tested. There's a lot left to do.
- src/cpp/RFC: other content, not ready for distribution.
- README.md: Documents the RFC7541 implementation process. At this point it's poorly written and contains a lot of muddled thinking. Once the writing is satisfactory it can be added with muddled thinking included.
- RFC7540.h/cpp: A placeholder for HTTP/2 functionality. It's currently only scaffolding and might not ever be used.
Make now also controls Python installation. Python 3.8 implemented PEP 517 and 518, changing the install mechanism. This updated mechanism is now used.
PyQt5, which is required by the Golf program, is not installed by default. PyQt5 requires qt5, which installs separately.
The stdio (Python C extention) installs using setuptools. The local Python library currently installs using sudo to install a link to the library subdirectory.
Fixed problems found in full recompile on Ubuntu test build machine running GCC 11.4.0 and ImageMagick-6. With ImageMagick-6, src/cpp/NN2 does not compile, so we changed its Makefile.BSD to reflect that fact but ONLY on that build machine. (Cygwin and Fedora's ImageMagick is version 7.) Fixed ~/src/cpp/Clone/FSlist.cpp, so it's no longer deprecated.
Removed links that cannot be resolved in the distributed content, retaining the associated "how to obtain it" information.
Created configure/build Makefiles and fixed most compile problems found in full recompile. (~/src/cpp/Clone/FSlist.cpp temporarily deprecated.)
Promoted Ioda from dev library into pub library.
Fixed documentation problems, primarily with regard to using characters "<" and ">".
(Primarily a documentation test release.)
This is a preliminary documentation release needed to verify that all the updated documentation is actually provided and not, for example, hidden by .gitignore.
The public domain license was changed from the "un-license" to the Creative Commons CC0 public domain license. This seems to be more worldwide in scope.
It does have a trade mark and patent restriction clause that might give pause to copiers, but the README.md "Copying" section essentially asserts that trade mark or patent restrictions do not and will not apply.
(Primarily a documentation release.)
On Windows/Cygwin, when running certain DEV library stress tests the Windows resource manager shows continual but small virtual storage growth, which could/should indicate a memory leak. The equivalent Linux top command does not show any storage growth. Storage allocation and release trace hooks were added, but the memory leak is yet to be found.
(A synchronization commit with debugging hooks left in the code. Most of these hooks need to be removed before a trunk commit.)
Issue #2 has been corrected, but dispatch::Task::~Task still has issues. The FC_UNDEF wait request doesn't complete, so the destructor doesn't complete and the Client and Server destructors don't complete. (Client and Server objects contain dispatch::Task objects.)
Isolating the the problem to the dispatch::Task destructor proved exceedingly difficult and compounded by the FC_UNDEF wait request (lack of) completion issue. Minor ancillary code base modifications were made while waiting for some flash of inspiration for ways to debug the problem.
This commit synchronizes the local and remote maint branch. This makes it easier to focus on the FC_UNDEF processing issue.
Distribution test: Moved prerequisite build programs to ~/src/cpp/sys/.
Branch trunk merged with maint.
Includes updated java golfer files and miscellaneous copyright updates, correcting either typos, formatting, or missing information.
Distribution test: ~/src/java/Webapp/*
Distribution test: ~/src/java/Webapp/*