Before version 4.3, commit 3eab405e, nano would write a help text to
a temporary file and then read this file into a buffer. The NOREAD
flag interfered with the latter, so needed to be reset for showing
help. But since the mentioned commit nano writes a help text directly
into a buffer and the unsetting of the NOREAD flag is superfluous.
The display code never references the LINE_NUMBERS flag -- the trigger
for displaying line numbers in front of the text is 'margin > 0'. So,
since commit 90bd25c1 that put help texts into a "normal" buffer, the
unsetting of the LINE_NUMBERS flag has been superfluous.
As a special case, and because it doesn't need any feedback on the
status bar nor other calls, add a special function and key binding
to allow the user to flip the LINE_NUMBERS flag also in a tiny nano.
This fixes https://savannah.gnu.org/bugs/?61441.
Reported-by: Hannu Nyman <hannu.nyman@iki.fi>
When the terminal has just one row, there is no room to constantly show
the position of the cursor, so do not allow the user to switch it on.
This fixes https://savannah.gnu.org/bugs/?61445.
Bug existed since version 5.8, commit 36ffb5f0.
With --zero or 'set zero', nano will hide the title bar or the minibar
(whichever is active) and will use all rows of the terminal for showing
the text of the buffer. Only when there is an important message will
it be shown on the bottom row -- until the next keystroke.
This feature can be toggled with M-Z.
The feedback at startup ("Read nnn lines") is suppressed with --zero,
because it disrupts the fullscreen experience (and is hardly useful).
The "Reading..." message then needs to be suppressed too, otherwise
it creates an annoying little flash.
There are still some issues that need to be addressed:
* In the browser, when the highlighted item is on the bottom row, it
will get obscured by the prompt bar or the "Search Wrapped" message.
* In the edit window, when the search occurrence is on the bottom row,
it can get obscured by the "This is the only occurrence" message.
* When tabbing at a prompt shows possible completions, they are shown
a row too low.
The text to be justified has been excised from the buffer and is now
in the cutbuffer, so we cannot compare any of its lines to 'filetop'.
This fixes https://savannah.gnu.org/bugs/?61438.
Bug existed since version 4.0, commit 14c08589.
In the beginning, the goal for nano was to be a drop-in replacement for
Pico. Pico did not know a Suspend command, so the ^Z keystroke needed
to be conditionalized on an option (-z or --suspend or the toggle M-Z),
just like the ^S and ^Q keystrokes (for stopping and resuming terminal
output) were conditionalized on --preserve.
But nano has abandoned full Pico compatibility since version 4.0. It
is time to unconditionalize ^Z as well. This should not be a problem:
Debian and Ubuntu have had 'set suspend' in their /etc/nanorc for years,
so a considerable portion of nano users have had ^Z enabled by default
for a long time, and no one seems to have complained.
If the keystroke bothers some user, they can unbind it in their nanorc.
They will still be able to suspend nano through the Execute menu: ^T^Z.
This addresses https://savannah.gnu.org/bugs/?61372.
The spotlighting should only be dropped when in the main menu, like in
the code thirty lines up, where 'timed' became TRUE only for MMAIN.
This fixes https://savannah.gnu.org/bugs/?61398.
Bug existed since version 5.8, commit 3f340836.
Normally, returning to the main loop will set 'focusing' back to TRUE,
but the replacement loop doesn't return until replacing is finished.
This fixes https://savannah.gnu.org/bugs/?61397.
Bug existed since version 5.0, commit d8249917.
When --breaklonglines is in effect and the user pasted just a few words
(anything without a linebreak), then act as if this short text had been
typed by the user and hard-wrap the line when it became overlong.
This fulfills https://savannah.gnu.org/bugs/?61353.
When --wordbounds (-W) is active, nano considers punctuation as
word-forming characters and will thus count words the same way
as 'wc -w' does. This is how nano counted words until now.
But when --wordbounds is not active (the default), only letters
and digits will be considered word-forming and thus lone groups
of lines and dashes and other punctuation will not be counted
as words, which is more like how a human would count words.
This addresses https://savannah.gnu.org/bugs/?61367.
In olden times, each help text started with a title line in the text
window followed by a blank line. But since version 2.8.2, since the
help texts have become almost regular buffers (and thus searchable),
the title of the help text is in the title bar, and since version 4.0,
since --morespace became the default, the text will start immediately
below it. But a title line immediately followed by text, without a
blank line between them, does not look nice. So, add such a blank
line back when not using --emptyline (and also when using --minibar,
because the top of the terminal window is like a title bar).
When the history file has been created by nano, it will not contain
any duplicate search or replace strings, nor duplicate commands, so
checking for such a duplicate for each read item was a waste of time.
And if the user edited the history file and created duplicates, who
are we to filter them out? They will not cause the history mechanism
to malfunction; they just take a little extra memory.
(Aaah, being able to define a keyboard macro
came in very handy when making this edit.)
This fixes https://savannah.gnu.org/bugs/?61331.
Bug existed since before version 2.2.0.
If all is programmed well, then the functions 'get_older_item' and
'get_newer_item' will only occur when a history list is passed to
acquire_an_answer(). But let's keep it certain we don't dereference
a NULL pointer.
When the user immediately cancels a search (^W^C), then nothing in
the history stack has changed, so then there is no need to reset the
history pointer to the bottom.
This slightly improves the fix for https://savannah.gnu.org/bugs/?61316.
This also gets rid of an assignment in an 'if' clause (twice),
elides a local variable, and makes it clearer that a pointer
gets moved to the previous or next item (instead of hiding it
as a side effect of the function call).
Whenever invoking a search prompt, calling up historical search items
should start with the newest item.
This fixes https://savannah.gnu.org/bugs/?61316.
Bug existed since version 2.8.2, since help texts became searchable.
When the answer at the prompt was clear before the user went up into
the history list, the answer should be cleared again when the user
comes back down to the present.
This fixes https://savannah.gnu.org/bugs/?61308.
Bug existed since before version 2.2.0.
For consistency, these five names ought to start with "flip_", but
then the bindable functions would need to be renamed too, and that
would be annoying for the user, *and* it would create too many names
that start with "flip" -- it wouldn't be distinctive any more.
When the current filename is equal to the filename of the buffer,
then clear the filename, because the chances that the user wants
to append or prepend a file to itself are practically nil.
This fulfills https://savannah.gnu.org/bugs/?61243.
The keystrokes are not listed, but should work like
they do in a normal, editable buffer.
This fixes https://savannah.gnu.org/bugs/?61274.
Bug existed since version 4.0, commit bc98dbca.
The functions to_first_file() and to_last_file() can get called from
do_statusbar_input(), which is called indirectly from do_prompt(),
and are expected to make the corresponding adjustment.
This fixes https://savannah.gnu.org/bugs/?61273.
Bug existed since version 5.0, commit 07c1ac90.
This avoids an unused and misleading assignment that might make
someone think the incremented variable will be used again.
Signed-off-by: Hussam al-Homsi <sawuare@gmail.com>
The calls of write_file() in do_spell() and do_formatter() write out
a temporary file -- such a file does not have an 'openfile' record,
so setting 'annotate' to TRUE was pointless. Using the correct value
for this parameter allows dropping two conditions in write_file().
And when 'exiting' is TRUE and --save-on-exit is in effect, it is
slightly wasteful to stat() the written file and store the info in
its 'openfile' record as the buffer will be closed directly after.
So, now the only place where 'annotate' is TRUE is when writing out
the full buffer to a normally named file.
First, we don't want the writing of an emergency file to be interrupted
by the user. But more important: the routine for restoring the handler
also disables SIGINT, which would leave the terminal with a non-working
Ctrl+C.
Saving an emergency file calls write_file() in a unique manner: with
thefile == NULL, fullbuffer == FALSE (even though the entire buffer
will be saved, of course) and tmp == TRUE (even though it is not a
temporary file, as it will persist after nano exits). But in fact
we want the handler for Ctrl+C installed only for normal files, not
for temporary files and not for emergency files -- the user should
not be able to interrupt the writing of those.
This fixes https://savannah.gnu.org/bugs/?61237.
Bug existed since version 4.3, commit 8550c6bd.
Trying to prepend would hang nano until some other process writes
something to the fifo.
This fixes https://savannah.gnu.org/bugs/?61236.
Bug existed since before version 2.2.4.
Curses cannot function if not informed via TERM which type of terminal
is being used. As many terminals are mostly compatible with a VT220,
falling back to "vt220" when TERM is unset has a good chance of giving
the user a usable nano, instead of simply failing.
(Falling back to "vt100" is not good as it contains padding delays.)
This partially addresses https://bugs.debian.org/991982.
In the terminal font that I use nowadays (that deals well with
combining characters), the normal diamond exceeded the boundaries
of a single cell and thus partly covered the first character of
the relevant line.
When a segment of text is extracted, copy any anchors that it has
into the cutbuffer, so that they get saved in the undo item, so that
an undo will put these anchors back on the lines where they were,
instead of leaving an inherited anchor at the top of the segment.
And when text is pasted, clear any anchors in it, so that they will
not travel nor multiply.
This fixes https://savannah.gnu.org/bugs/?61162.
Bug existed since version 5.0, since anchors were introduced.
That is: pause a little to jolt the user, and then leave the message
on the status bar.
This fixes https://savannah.gnu.org/bugs/?61147.
Bug existed since version 4.8, commit 416386ed.
This fixes https://savannah.gnu.org/bugs/?61146
Bug existed since version 2.3.2, since --locking was introduced,
and continued to exist since version 4.8 despite commit 497f126b
claiming to have fixed it.
The documentation (https://www.haiku-os.org/docs/api/libroot.html)
says that BSD extensions are enabled by default, but somehow this
does not seem to be the case on my system. With _DEFAULT_SOURCE
defined, though, src/files.c now compiles fine.
When safe_tempfile() returns a valid filename, it should also
return a valid open stream.
This fixes https://savannah.gnu.org/bugs/?61064.
Bug existed since version 1.3.8, commit 5e068c60.
Any error message will already have been written to the terminal by
statusline(). Also, "Buffer not written to" looked too much like
"Buffer written to"; one had to look closely to see the difference.
Make sure that when multiple nanos with nameless buffers die,
each of them tries saving those buffers to different files.
This addresses https://savannah.gnu.org/bugs/?60902.
Reported-by: Julian Rosen <julianrosen@gmail.com>
Having a local variable that gets assigned to just once looks rather
poor. To me it makes more sense to have two separate exit points: one
for failure somewhere in the middle, and one for success at the end.
This way it is completely clear which value gets returned when.
Benno Schulenberg