| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
* should also fix Win32 nightly builds
* Even though we weren't using main's argv, but were using glib
API for retrieving the command line in UTF-8, newer MinGW runtimes
would fail when converting the Unicode command line into the system codepage
would be lossy.
* Most people seem to compile in a "manifest" to work around this issue.
But this requires newer Windows versions and using some Microsoft tool which isn't
even in $PATH.
Instead, we now link with -municode and define wmain() instead, even though we still
ignore argv. wmain() proabably get's the command line in UTF-16 and we'd have to
convert it anyway.
* See https://github.com/msys2/MINGW-packages/issues/22462
|
|
not found' error
* This is important with gotos in loops as in <@O/x/> where, we would otherwise
get a confusing "Unterminated loop" error.
* This in particular fixes the error thrown in grosciteco.tes when encountering
a new unknown command.
|
|
* The program exit code will usually not signal failures since they are caught earlier.
* Therefore, we always have to capture and check stderr.
|
|
characters in Q-Register)
* It was initialized only once, so it could inherit the wrong local Q-Register table.
A test case has been added for this particular bug.
* Also, if starting from the profile (batch mode), the state machine could be initialized
without undo, which then later cause problems on rubout in interactive mode.
For instance, if S^EG[a] fails and you would repeatedly type `]`, the Q-Reg name could
grow indefinitely. There were probably other issues as well.
Even crashes should have been possible, although I couldn't reproduce them.
* Since the state machine is required only for the pattern to regexp translation
and is performed anew for every character in interactive mode,
we now create a fresh state machine for every call and don't attempt
any undo.
There might be more efficient ways, like reusing the string building's
Q-Reg parser state machine.
|
|
* A test case has been added, although it might have been accidental
that on caused crashes.
|
|
|
|
* You can now specify `--with-scitecodatadir` as a relative path,
that will be interpreted relative to the binary's location.
* Win32 binaries already were relocatable, but this was a Windows-specific
hack. Win32 binaries are now built with `--with-scitecodatadir=.`
since everything is in a single directory.
* Ubuntu packages are now also built `--with-scitecodatadir=../share/sciteco`.
This is not crucial for ordinary installations, but is meant for AppImage creation.
* Since AppImages are now built from relocatable packages,
we no longer need the unionfs-workaround from pkg2appimage.
This should fix the strange root contents when autocompleting in
AppImage builds.
* This might also fix the appimage.github.io CI issues.
I assume that because I could reproduce the issue on FreeBSD's
Linuxulator in dependence of pkg2appimage's "union"-setting.
See https://github.com/AppImage/appimage.github.io/pull/3402
* Determining the binary location actually turned out be hard and
very platform-dependant. There are now implementations for Windows
(which could also read argv[0]), Linux and generic UNIX (which
works on FreeBSD, but I am not sure about the others).
I believe this could also be useful on Mac OS to create app bundles,
but this needs to be tested - currently the Mac OS binaries are
installed into fixed locations and don't use relocation.
|
|
string cells
Found thanks to the "infinite monkey" test.
|
|
Supposing that any monkey hitting keys on a typewriter, serving as a hardcopy
SciTECO terminal, will sooner or later trigger bugs and crash the application,
the new monkey-test.apl script emulates such a monkey.
In fact it's a bit more elaborate as the generated macro follows the frequency
distribution extracted from the corpus of SciTECO macro files (via monkey-parse.apl).
This it is hoped, increases the chance to get into "interesting" parser states.
This also adds a new hidden --sandbox argument, but it works only on FreeBSD (via Capsicum)
so far. In sandbox mode, we cannot open any file or execute external commands.
It is made sure, that SciTECO cannot assert in sandbox mode for scripts that would
run without --sandbox, since assertions are the kind of things we would like to detect.
SciTECO must be sandboxed during "infinite monkey" tests, so it cannot accidentally
do any harm on the system running the tests.
All macros in sandbox mode must currently be passed via --eval.
Alternatively, we could add a test compilation unit and generate the test data
directly in memory via C code.
The new scripts are written in GNU APL 1.9 and will probably work only under FreeBSD.
These scripts are not meant to be run by everyone.
|
|
failed, get assigned to error frames
|
|
jumping to the beginning of the macro)
* I am not sure whether this feature is really that useful...
* teco_machine_main_t::macro_pc is now pointing to the __next__ character to execute,
therefore it's easier to manipulate by flow control commands.
Also, it can now be unsigned (gsize) like all other program counters.
* Detected thanks to running the testsuite under Valgrind.
|
|
* There was some boilerplate code missing in teco_state_search_all_initial(),
that is present in teco_state_search_initial().
* Perhaps there should be a common function to avoid redundancies?
* This will also fix the initialization of the string argument codepage for <N>.
|
|
* Has been observed on PDCursesMod/WinGUI when pressing CTRL+Shift+6 on an US layout.
I would expect code 30 (^^) to be inserted, instead PDCurses reports two keypresses (6^^).
The first one is now filtered out since this will not be fixed upstream.
See also https://github.com/Bill-Gray/PDCursesMod/issues/323
* Since AltGr on German layouts is reported as CTRL+ALT, we must be careful not to filter those
out as well.
* This is active on all PDCurses variants - who knows which other platforms will behave similarily.
* You still cannot insert code 0 via CTRL+@ since PDCurses doesn't report it, but ncurses does not allow that either.
This _could_ be synthesized by evaluating the modifier flags, though.
|
|
* Fixes the test suite on PDcurses/Win32 and therefore CI builds.
* Should be necessary on UNIX as well since later on, we would access
cmdline_window, which is not yet initialized.
I didn't see any errors in Valgrind, though.
|
|
* Supports all immediate editing commands.
Naturally it cannot emulate arbitrary key presses since there is no
canonic ASCII-encoding of function keys.
Key macros are not consequently also not testable.
The --fake-cmdline parameter is instead treated very similar to
a key macro expansion.
* Most importantly this allows adding test cases for rubout behavior
and bugs that are quite common.
* Added regression test cases for the last two rubout bugs.
* It's not easy to pass control codes in command line arguments in
a portable manner, so the test cases will often use { and }.
Control codes could be used e.g. by defining variables like
RUBOUT=`printf '\b'`
and referencing them with ${RUBOUT}.
|
|
* This was a regression introduced in 41ab5cf0289dab60ac1ddc97cf9680ee2468ea6c,
which changed the semantics of teco_doc_undo_set_string().
* Removed undo_append_string() Q-Reg virtual method.
append_string() now does its own undo token emission, so that we can
defer the teco_doc_undo_edit() after the point that the document
was initialized. This is important, so that we can configure the
default encoding on new registers.
|
|
* detected under FreeBSD
* It turns out that it's unsafe to make the GIOChannel blocking
even though the application has already terminated and the channel
should be closed automatically.
The channel does not report EOF, but instead we have to look for
zero reads - in complete contrast to the behavior on Windows.
Apparently, it's very tricky to use this API correctly
(ie. it sucks).
* fixup to e9bef20a8ad89d304fe3e8fafa00056d22de2326
|
|
* We now recreate the event loop with every call since
it turned out that the idle watcher wouldn't be invoked
after the event loop has been quit once.
This at least fixes interruption of ECbash -c 'while true; do true; done'$.
* Unfortunately, ECping -t 8.8.8.8$ still cannot be interrupted
(unless you manually kill the process from the task manager).
|
|
* This API behaves very strangely and differently compared to UNIX/X11.
When getting, it returns a trailing null for all clipboard contents
(unless the clipboard is empty) and when setting, we apparently have to include it as well.
At least since we cut it off when getting.
Even more strangely, setting without the trailing null did work
when pasting in external apps. (How they know when it's safe to throw
away the trailing null is mysterious.)
* In other words, this fixes X~G~.
|
|
* This has only ever observed on Win32, probably because the spawning
behaves very differently.
* The stdout watcher could be invoked even after removing the source,
so we must be secured against it - this was causing some overflows
and invalid reads.
* Also, teco_eol_reader_convert() could return 0 even after process
termination, which would sometimes result in too few bytes being
inserted.
This could be provoked relatively easily by invoking ECdir$ repeatedly.
|
|
* Contrary to the Gtk documentation, the gtk_selection_data_get_length()
already includes a trailing null, so we always inserted a bogus
null char when using G~ or ^EQ~.
|
|
* The old bug of saving gchar in gints, so teco_eol_reader_t::last_char could become negative.
* When converting from an UTF-8 text with CRLF linebreaks, we could have data loss and corruptions.
* On strings ending in UTF-8 characters, teco_eol_reader_t::offset would overflow, resulting
in invalid reads and potentially insertion of data garbage.
I observed this with G~ on Gtk.
* Test cased updated. Couldn't reproduce the bug with the test suite, though.
|
|
terminating process groups on Win32
* Sometimes already the job assignment failed in CI builds.
We now check whether the process is still alive before throwing an error.
* We now set the JOB_OBJECT_LIMIT_KILL_ON_JOB_CLOSE flag.
This theoretically shouldn't be necessary when using TerminateJobObject(),
but who knows.
|
|
* The previous check could result in false positives if you are editing
a local Q-Register, that will be destroyed at the end of the current macro frame,
and call another non-colon modified macro.
* It must instead be invalid to keep the register edited only if it belongs to the
local Q-Registers that are about to be freed.
In other words, the table that the currently edited Q-Register belongs to, must be
the one we're about to destroy.
* This fixes the solarized.toggle (F5) macro when using the Solarized color scheme.
|
|
This reverts commit 024d26ac0cd869826801889f1299df34676fdf57.
This was re-introducing Clang warnings since gboolean is signed.
I should have read the git blame before re-introducing gboolean...
|
|
current state machine
* The previous solution was not wrong, but unnecessarily complex. We already have a flag
for exactly this purpose.
* Avoid redundancies by introducing teco_machine_stringbuilding_set_codepage().
|
|
teco_machine_stringbuilding_t::codepage
* It's contained in teco_machine_main_t which is created per macro call frame.
So after macro calls, the machine no longer exists.
It is therefore unsafe to undo its members indiscriminately.
* On the other hand, we must undo the codepage setting when run interactively,
so it is now only undone when belonging to the commandline macro frame.
* This was actually causing memory corruptions on every fnkeys cursor movement, but never
caused crashes - probably because the invalid pointers are always pointing to unused
parts of the C call stack.
* Initially broken in b31b8871.
|
|
* this would also leak a few bytes on every of fnkeys.tes' movement commands
|
|
* Apparently, netbsd-curses overwrites the escdelay on initscr() (if $ESCDELAY is not set),
so we have to apply the default 25ms after screen initialization.
* The info line is not drawn correctly on netbsd-curses, but only on st/simpleterm.
I assume this is just a shortcoming of the included terminfo entry.
|
|
allow breaking from within braces
For instance, you can now write <23(1;)> without leaving anything on the stack.
|
|
characters after escaped characters
|
|
* added TECO_ERROR_CLIPBOARD for all clipboard-related errors
|
|
* makes it possible, albeit cumbersome, to escape pattern match characters
* For instance, to search for ^Q, you now have to type
S^Q^Q^Q^Q$.
To search for ^E you have to type
S^Q^Q^Q^E$.
But the last character cannot be typed with carets currently (FIXME?).
For pattern-only characters, two ^Q should be sufficient as in
S^Q^Q^X$.
* Perhaps it would be more elegant to abolish the difference between string building
and pattern matching characters to avoid double quoting.
But then all string building constructs like ^EQq should operate at the pattern level
as well (ie. match the contents of register q verbatim instead of being interpreted as a pattern).
TECOC and TECO-64 don't do that either.
If we leave everything as it is, at least a new string building construct should be added for
auto-quoting patterns (analoguous to ^EN and ^E@).
|
|
|
|
|
|
|
|
* Previously you could open files of arbitrary size and the limit would be checked only afterwards.
* Many, but not all, cases should now be detected earlier.
Since Scintilla allocates lots of memory as part of rendering,
you can still run into memory limits even after successfully loading the file.
* Loading extremely large files can also be potentially slow.
Therefore, it is now possible to interrupt via CTRL+C.
Again, if the UI is blocking because of stuff done as part of rendering,
you still may not be able to interrupt the "blocking" operation.
|
|
* We now set opt.retain=false for the process, so jemalloc returns
freed memory and the RSS decreases when recovering from memory limit hits.
This should be safe at least on FreeBSD.
* Either the opt.retain option is new or I was previously testing
this only on 32-bit systems.
|
|
overflow checking
* teco_memory_usage is now an unsigned integer.
* Unfortunately we currently rely on the variable being int-sized since we use
atomic operations.
This means on 64-bit systems, limiting will not work as expected if you set the limit larger
than 4GB.
Not sure whether this should be fixed.
* Calling teco_memory_check() with a non-null request-size was totally broken and could
result in bogus failures.
This is currently used exclusively for checking backwards searches.
|
|
* Curses: "icons" have also been added
|
|
|
|
command line argument
* For instance, you can now rub out ^Q^W at the beginning of a string argument.
Otherwise, pressing Ctrl+W after ^Q^W would rub out only the ^W.
The next Ctrl+W would then insert ^W, due to special immediate editing inhibition after ^Q.
* This still only works if the string building construct expanded to at least one byte.
Suppose you have ^EQq, expanding to nothing, pressing Ctrl+W would chain to the default
teco_state_process_edit_cmd() and the entire command would be rubbed out.
This is probably tolerable.
|
|
* This allows you to type ^Q^U (which would otherwise rub out the entire argument)
and ^Q^W (which would otherwise rub out the ^Q).
* ^Q^U coincidentally worked previously since the teco_state_stringbuilding_escaped
state would default to teco_state_process_edit_cmd().
But it's better to make this feauture explicit.
* This finally makes it possible to insert the ^W (23) char into a buffer.
In interactive mode, you can still only type Caret+W as a string building construct.
* ^G could also be inhibited after ^Q, but the control char is not used anywhere yet,
so there is no point in doing that.
|
|
* The XTerm version is still checked if we detect running under XTerm.
* Actually, the XTerm implementation is broken for Unicode clipboard contents.
* Kitty supports OSC-52, but you __must__ enable read-clipboard.
With read-clipboard-ask, there will be a timeout.
But we cannot read without a timeout since otherwise we would hang indefinitely
if the escape sequence turns out to not work.
* For urxvt, I have hacked an existing extension:
https://gist.github.com/rhaberkorn/d7406420b69841ebbcab97548e38b37d
* st currently supports only setting the clipboard, but not querying it.
|
|
* This is especially important on platforms, requiring the wgetch()
poll workaround to detect CTRL+C (PDCurses/WinGUI).
wgetch(cmdline_window) would implicitly wrefresh(cmdline_window),
which resulted in additional flickering when pressing function keys.
This is no longer so important since key macros are processed
as an unity and the cmdline will be updated only after processing
all of the characters contained in them, ie. only once after the key press.
Still, there could have still been unwanted side effects.
At the very least, wgetch(input_pad) should be faster.
* The XTerm clipboard implementation was getch()ing on stdscr,
so potentially suffered from the same problem.
It should be tested again.
* Since keypad() is now always enabled even on netbsd-curses.
I assume that the function key processing bug in netbsd-curses
has been fixed by now. We are not building any releases with
netbsd-curses. But it should be retested.
* It does not resolve all flickering issues on PDCurses/WinGUI.
Both the command line and the Scintilla view still flicker near
the cursor. See
https://github.com/Bill-Gray/PDCursesMod/issues/322
|
|
* This is necessary to fix the Unicode test suite on Win32,
so I was always passing in --disable-shared manually.
It's easy to forget though when building from scratch.
* We don't currently install any (shared) library, so this is safe
on all platforms.
In fact on all other platforms, libtool detects that and doesn't
generate wrapper binaries in any way.
Only on win32 it's apparently buggy.
|
|
* Turns out that "%C" in wprintw() does not work with non-ANSI chars.
* We still don't want to introduce the Curses widechar API,
so I added teco_curses_add_wc() as a replacement for wadd_wch().
|
|
characters
* Some characters like LF wouldn't be displayed in the message line correctly.
* In fact the Gtk UI cannot display any of the control characters correctly.
* I was considering deferring all echoing/formatting to the UIs, so they can use
TecoGtkLabel or teco_curses_format_str().
This is not possible since messages transmitted via GError must not contain null-bytes,
so these need to be sorted out earlier anyway.
* This should also fix syntax errors in PDCurses for Windows where "%C" apparently doesn't
work with non-ANSI codepoints.
|
|
all expansions of ^EQq, ^EUq and so on
* Previously, there was no way to enter upper-case mode in interactive commands since
the Ctrl+W immediate editing command is interpreted everywhere.
* Without the case folding of ^EQq/^EUq results, the upper and lower case modes are actually pretty useless
considering that modern keyboards have caps lock.
So it was clear we need this, regardless of what the classic TECOs did.
The TECO-11 manual is not very clear on this.
tecoc apparently does not case-fold ^EQq results.
* This opens up new idioms, for instance
`EUq^W^W^EQq$` in order to upper case register q.
It's also the only way you can currently upper-case Unicode codepoints.
|
|
* Ctrl+^ (30) and Caret+caret (^^) were both translated to a single caret.
While there might be some reason to keep this behavior for double-caret,
it is certainly pointless for Ctrl+^.
* That gives you an easy way to insert Ctrl+^ (code 30) into documents with <I>.
Perviously, you either had to insert a double-caret, typing 4 carets in a row,
or you had to use <EI> or 30I$.
* The special handling of double-caret could perhaps be abolished altogether,
as we also have ^Q^ to escape plain carets.
The double-caret syntax is very archaic from the time that there was no proper
^Q as far as I recall correctly.
|
