The 'openfile->fmt' element gets initialized to 'UNSPECIFIED',
so the code has to take that possibility into account.
This fixes https://savannah.gnu.org/bugs/?65676.
Bug existed since version 8.0, commit fe4f74f6.
This closes a window of opportunity where the emergency file could be
replaced by a malicious symlink.
The issue was reported by `MartinJM` and `InvisibleMeerkat`.
Problem existed since version 2.2.0, commit 123110c5, when chmodding
and chowning of the emergency .save file was added.
When recording a macro over a laggy connection or on a slow, overloaded
computer, several keystrokes could be recorded in one burst, and if any
but the last of those keystrokes was the shortcut for `runmacro`, then
running the macro would lead to an infinite loop and nano would hang.
This new implementation of snipping the "last keystroke" will, however,
snip *too many* keystrokes when several of them were recorded at once
and `runmacro` or `recordmacro` was among them, resulting in a cropped
and thus misrecorded macro. But... that's better than hanging.
In general, though, the user should be slow and deliberate when
recording a macro: waiting for nano to have processed the last
keystroke before typing the next.
This fixes https://savannah.gnu.org/bugs/?65649.
The issue was reported by `correctmost`.
Problem existed since version 2.9.0, since macros were introduced.
These calls of do_delete() were meant to delete just one character,
but over time do_delete() morphed into doing also other things...
Change the calls to invoke the correct function instead.
(This also avoids snipping any zero-width characters that come after
a snipped space, as that is probably not what the user wants.)
This fixes https://savannah.gnu.org/bugs/?65636.
The issue was reported by `correctmost`.
Bug existed since version 3.2, commit ae3ec178,
when --zap was introduced.
Redoing an automatic hard-wrap while one or more chunks of the affected
line are offscreen, can leave 'firstcolumn' with a value that doesn't
fit the situation. So, make sure that it has a valid value.
This complements the previous commit.
This fixes https://savannah.gnu.org/bugs/?65611.
The issue was reported by `correctmost`.
Bug existed since version 2.8.6, commit e375995d.
When one or more chunks of the current line are above the viewport,
and this line gets hard-wrapped, then 'edittop' needs to be advanced,
otherwise the first row could be blank -- representing a chunk that
doesn't exist any more.
This fixes https://savannah.gnu.org/bugs/?65604.
The issue was reported by `correctmost`.
Bug existed since version 2.8.6, commit e375995d.
When opening a nonexistent file with nano, it likely consists of only a
name without any path component, and thus without any slash. So when a
file regex checks for a slash, it should check also for start-of-string.
This fixes https://savannah.gnu.org/bugs/?65591.
Problem existed for the Makefile since version 2.9.8, commit 22663f8a,
and for .profile since version 3.0, commit 4a268678 (but earlier, nano
did not recognize .profile files at all).
When an entirely blank line is trimmed (when --autoindent is active and
Enter is pressed while the cursor is at the end of the current indent),
the mark needs be adjusted *before* 'current_x' is zeroed, otherwise the
mark gets moved too much to the right, which causes the region to become
bigger than what the user intended, or leads to accessing unallocated
memory.
This fixes https://savannah.gnu.org/bugs/?65586.
The issue was reported by `correctmost`.
Bug existed since version 2.8.1, commit 005ee8ed.
A "Read xx lines (converted from...)" message should be suppressed when
--zero or --minibar are active (just like the normal "Read xx lines"),
as otherwise it gets shown at a confusing moment when multiple files
are opened at the same time. The message should get shown, however,
when inserting a file into the current buffer.
This fixes https://savannah.gnu.org/bugs/?65560.
Bug existed since version 6.0, commit f147131e.
The idea of the warning was that `row < 0 || row >= editwinrows` would
never be true: that the check was redundant and could be removed.
As it appears to be too hard to make sure in advance that a chunk will
not fall outside the viewport, just drop the warning and leave the check
in place.
This also addresses https://savannah.gnu.org/bugs/?64168.
Reported-by: Matteo Raso <matteo_luigi_raso@protonmail.com>
When we're on the last row in the viewport, it is unwise to try and draw
the next line.
This fixes https://savannah.gnu.org/bugs/?65539.
The issue was reported by `correctmost`.
Bug existed since commit c2322f85 from three days ago.
Drawing a line will allocate multidata for it (even when the line is
actually empty), so that this multidata will not be missing later on
when the next line gets drawn in the same burst of keystrokes.
(This also removes the triggering of a full refresh when line numbers
are active: only the new magic line needs to be additionally drawn, in
order to get the line number printed.)
This fixes https://savannah.gnu.org/bugs/?65516.
The issue was reported by `correctmost`.
Problem existed since version 6.3, commit 80c2000f.
It is bindable only in the Insert menu -- in the Execute menu it has
a "blind", unadvertised binding (because I consider toggling between
menus somewhat of a misfeature).
On my current laptop, typing <Alt+Insert> is awkward because it requires
holding down <Fn> too. Also, M-" and M-' are in the normal symbols area
of the keyboard, which makes them easier to type and more discoverable.
Furthermore, being next to M-: and M-; (on a US keyboard) reinforces the
meanings: start/place, stop/remove, run/goto.
Normally, when recording a macro, users will make their keystrokes
slowly and carefully, and will most likely wait to see the effect
of the previous keystroke before making the next. So, the chances
of two `recordmacro` keystrokes coming in in quick succession is
normally nil. The 'macro_length' variable just needs a guard to
prevent it from underflowing when someone is hammering the keys.
This fixes https://savannah.gnu.org/bugs/?65394 in a better way.
When the justified region fits in its entirety onscreen, then it should
be shown in its entirety, because the user is probably curious what the
region looks like after justification. This behavior will mean that
the line above the topline of the screen will still have multidata
(when the syntax has multiline regexes). When the justified region
does not fit onscreen, then as much as possible should be shown --
meaning that the cursor should be either at the top or the bottom
of the viewport, meaning that it should not be centered.
This fixes https://savannah.gnu.org/bugs/?65482.
The issue was reported by `correctmost`.
Bug existed since version 4.0, since justifying a region was introduced.
When justifying a region that was marked backwards, the cursor
should stay at the beginning of this region. The old logic was
faulty because mark_is_before_cursor() does not give the correct
result *after* the justification has been done: for some reason
the mark ends up always before the cursor.
Bug existed since version 5.2, commit 0f8423eb,
which mistakenly removed the auxiliary variable. :/
The strings are gettextized further down, for the non-tiny version,
so they will get translated anyhow. The relevant translator hint
is the earlier one about "the next thirteen strings".
Also adjust an indentation, and ungettextize another string for
consistency.
With NetBSD curses, when only the cursor is moved (without writing any
text), then a call of wnoutrefresh() is needed to make doupdate() move
the cursor. Ncurses does not need this.
This addresses https://savannah.gnu.org/patch/?10438.
Benno Schulenberg