Also, avoid calling a case-changing function when it's not needed.
And elide a default case that can never occur (because the function
always gets called with one of the four letters).
Nano does not do anything for the keystroke anyway (KEY_B2, the center
key on the numeric keypad, gets translated to ERR later on), so there
is no point in recognizing some of its escape sequences -- let those
sequences result in "Unknown sequence" instead.
When --rawsequences is used, consume the full escape sequences for
the F17 to F24 function keys (on xterm-compatible terminals and on
the Linux console), to prevent them entering junk into the buffer.
This fixes https://savannah.gnu.org/bugs/?57483.
Go back to how line numbers were colored in versions 2.7.1 to 4.3 --
coloring the space (when a background color is used) was unintented.
This fixes https://savannah.gnu.org/bugs/?57412.
Bug existed since version 4.4, commit 699cacf7.
It allows entering a TAB character at the cursor position while the
mark is on the same line as the cursor, as some users sometimes do.
This refines the feature that was added in version 2.9.2, and makes
it behave like in the Gedit and Kate editors, for example.
This addresses https://savannah.gnu.org/bugs/?57357.
Reported-by: Sébastien Desreux <seb@h-k.fr>
After the previous change, all remaining calls of parse_mbchar() have
NULL as their third parameter. So, drop that parameter and remove the
chunk of code that handles it. Also rename the function, as there are
already too many functions that start with "parse".
The old default behavior of showing the first three messages with a
long pause after each of them was annoying, and the final "Further
messages were suppressed" hid the relevant information. So, when
there is more than one error message, just pause very briefly and
then add trailing dots to the first message.
This makes the 'nopauses' option a no-op.
This addresses https://savannah.gnu.org/bugs/?57048.
In a non-UTF8 build, mbwidth() returns always 1, so it is pointless
to call that function and compare its result to zero then.
Also, don't bother special-casing the function for a non-UTF8 locale.
When there are fewer than four menu items, act as if there are four:
make their clickable width half the width of the screen.
Until now, when there were two menu items, their clickable width would
be the full width of the screen, which was overwide. But when there
was just one menu item, its clickable width would suddenly be as small
as when the menu had the maximum number of items (12 for an 80-column
screen). This was odd.
Also, slightly simplify another computation.
In this way the keycode cannot be unbound from the 'indent' function,
so pressing <Tab> on a marked region will always indent the region.
This fixes https://savannah.gnu.org/bugs/?56960.
Bug existed since the "tabbing" of a marked region was introduced,
in version 2.9.2, commit 09958ebd.
Make display_string() report back (via a global variable) whether the
just converted string has room to spare within the allotted span, so
that draw_row() can avoid trying to clear the remainder when there is
nothing to clear.
This fixes https://savannah.gnu.org/bugs/?56562.
Bug existed since commit 699cacf7 from three days ago.
It is a waste of time to first fully clear a row when right afterward
text will be written to it (for most of the row, on average). So...
just clear the part of the row after the written text. Curses has the
perfect function for this: clrtoeol().
Where the row needs to be filled with spaces with the attributes that
were set (title bar, prompt bar, and browser highlight), use printw().
But when the row needs to be actually cleared, use the much faster
clrtoeol().
This reduces the time needed to clear a full row with 95 percent.
Also, don't use mblen() directly, to not get stuck when it returns -1,
which it will when running a non-UTF-8 build in a UTF-8 locale.
This fixes https://savannah.gnu.org/bugs/?56472.
Bug existed since commit cd094822 from yesterday.
The majority of characters in the files that get edited with nano will
be single-byte printable ASCII characters, so their case should come
first in the main loop of display_string().
In addition, the function was used just once, had a weird return value,
and now some more code can be excluded from a non-UTF8 build.
Make use of the fact that any single-byte character always occupies
just one column, and call the costly mbtowc() and wcwidth() only for
characters that actually are multibyte.
Instead of calling in twenty places parse_mbchar(pointer, NULL, NULL),
use a simpler and faster char_length(pointer). This saves pushing two
unneeded parameters onto the stack, avoids two needless ifs, and elides
an intermediate variable.
Its main purpose will follow in a later commit: to speed up searching.
Add a temporary boolean for this, because isendwin() returns TRUE
only when curses mode has actually been started with initscr() and
then exited with endwin().
The extra things that close_buffer() did are only needed and useful
when manually closing a buffer, so move them there. The other three
calls of close_buffer() only need to get rid of the current buffer
(making the preceding buffer the new current one) and nothing else.
When a help-text buffer is closed while exiting from the help viewer,
the rest of the exit code takes care that the preceding buffer gets
displayed properly. And when a help-text buffer is closed because a
SIGWINCH occurred, doing anything for the preceding buffer is a waste
of time because a new help-text buffer will be opened and displayed
instantly.
Determine the actual number of bytes the striped character consists of,
instead of assuming it's simply one, and determine the real column that
the character starts in, instead of assuming it's the stripe column.
This fixes https://savannah.gnu.org/bugs/?55917
and fixes https://savannah.gnu.org/bugs/?55922.
Condensed-by: Benno Schulenberg <bensberg@telfort.nl>
Option -J (--guidestripe) takes a column number as argument and then
shows a vertical, colored bar over the entire height of the buffer,
to aid the user in regulating the width of the text when the terminal
is wider than this desired width.
This fulfills https://bugs.debian.org/916392.
Requested-by: Arturo Borrero González <arturo@debian.org>
And fulfills https://savannah.gnu.org/bugs/?55315.
Requested-by: Bryan Christ <bryan.christ@gmail.com>
When the right half of a two-column character gets overwritten by the
">" continuation character, then represent its left half as "[" also
when spotlighting a replacement candidate. Achieve this by calling
display_string() in the overshooting case with 'isprompt' set to TRUE,
which also makes this function reserve space for the ">" continuation
character so spotlight() doesn't need to do this any more.
This fixes https://savannah.gnu.org/bugs/?55679.
Improved-by: Benno Schulenberg <bensberg@telfort.nl>
The change in commit edc0d628 was flawed. The old logic covered
every case but one: having reached the end of 'buf' but 'column'
still having overshot the screen width. Now cover this latter
case too.
This fixes https://savannah.gnu.org/bugs/?55638 properly,
allowing the next commit to fix another buglet.
Condensed-by: Benno Schulenberg <bensberg@telfort.nl>
That is: keep nibbling off characters until a character is eaten that
takes up at least one column.
This fixes https://savannah.gnu.org/bugs/?55759.
Reported-by: David Lawrence Ramsey <pooka109@gmail.com>
Bug existed since version 2.8.0, commit 5283acdc.
To make it obvious that they are not actual characters in the file.
The default highlighting is in reverse video, but this can be changed
to bold by using --bold or 'set boldtext'.
This fulfills https://savannah.gnu.org/bugs/?55571.
The angular brackets look a bit like arrows and thus hint much more
at the idea of continuation than a dollar sign.
With-help-from: David Lawrence Ramsey <pooka109@gmail.com>
When the screen's edge leaves just one column for a two-column
character, then show its left half as "[" and its right half as "]".
(They used to be shown as ">" and "<", but these characters will be
repurposed in the next commit.)
This addresses https://savannah.gnu.org/bugs/?55657.
Reported-by: David Lawrence Ramsey <pooka109@gmail.com>
It needs to be trimmed and substituted with a single-width placeholder
to prevent the continuation character ">" from getting pushed to the
next row.
Also, use the correct width for the answer so that the continuation
character is only displayed when the answer does not fit any more.
This fixes https://savannah.gnu.org/bugs/?55620.
Reported-by: David Lawrence Ramsey <pooka109@gmail.com>
Original-patch-by: David Lawrence Ramsey <pooka109@gmail.com>
When a line ends with a two-column character, and this character straddles
the edit window's edge, then, although we've reached the end of 'buf', we
have exceeded the 'beyond' column and this two-column character needs to
be trimmed from the 'converted' string, otherwise it would get printed on
the next row.
This fixes https://savannah.gnu.org/bugs/?55638.
Bug existed since version 2.8.0, commit 5283acdc.
After a series of escapes, also <Shift+Meta+letter> should be recognized
as a command keystroke, not just <Meta+letter>.
This fixes https://savannah.gnu.org/bugs/?55442.
Bug existed since the ignoring of <Escape>s before a valid command
keystroke was introduced in version 3.0, commit ecc9211a.
When the terminfo selected by TERM does not match the terminal and
the <Delete> key behaves the same as <Backspace> (deleting leftward)
or the <Backspace> key behaves the same as <Delete> (deleting without
moving the cursor), then using just -d or --rebinddelete should make
the deviant key behave correctly again without affecting the other.
Set the Meta flag only for the regular Meta keystrokes: "M-" plus a
printable ASCII character. The special keycodes are unique and do
not need any extra flags to differentiate them.
This works for xterm, rxvt, Eterm, Konsole, and xfce4-terminal, which
generate "Esc [ 3 ; 3 ~", but not for urxvt, which generates one of
its double-escape sequences, "Esc Esc [ 3 ~".
When something is spotlighted, it should survive a refresh of
the screen and an excursion to a help text, so the spotlight
should get painted whenever the edit window is drawn.
This fully fixes https://savannah.gnu.org/bugs/?54721.
Ask ncurses for the codes for the Shift+arrow keys, so that also
<Shift+Up> and <Shift+Down> can be recognized, for which ncurses
doesn't have standard codes.
This fixes https://savannah.gnu.org/bugs/?54790.
Reported-by: Javier Valencia <javiervalencia80@gmail.com>
This allows running the speller (default binding: ^T) also on files for
which a linter has been defined. This makes it possible to spell check
comment blocks in source files, for example.
This fulfills https://savannah.gnu.org/bugs/?54711.
Requested-by: Mike Frysinger <vapier@gentoo.org>
On FreeBSD and NetBSD (when reached through ssh from a Linux machine)
this has the absurd effect of making <Ctrl+Backspace> do a 'cutwordleft'
by default, out of the box, without needing any rebindings. Weird, but
wonderful, because the ideal behavior.
Also ensure that <Shift+Delete> always does a Backspace.
This makes that we have the following set of "congruent" keys:
<Tab> moves text to the right,
<Shift+Tab> moves text to the left,
<Delete> "eats" a character to the right,
<Shift+Delete> "eats" a character to the left,
<Ctrl+Delete> "eats" a word to the right,
<Shift+Ctrl+Delete> "eats" a word to the left.
Since the last version, the user can filter an entire buffer through
an external command. This external command can also be a formatting
program, so there is no longer any need for this specific and special
formatter command.
Bind the until-now unbound function 'cutwordright' to <Ctrl+Delete>.
The complementary function, 'cutwordleft', is not bound by default
because on many terminals the keystroke <Ctrl+Backspace> generates
^H -- the canonical ASCII backspace character. We cannot change the
existing action of ^H without upsetting some users.
Signed-off-by: Marco Diego Aurélio Mesquita <marcodiegomesquita@gmail.com>
Signed-off-by: Benno Schulenberg <bensberg@telfort.nl>
When using option -K on an xterm-like terminal, pressing <Shift+Del>,
<Alt+Del>, or <Ctrl+Del> would enter "2~", "3~", or "5~", respectively,
into the buffer. Now it will just report an "Unknown sequence".
Just like an <Esc> before a Ctrl+letter keystroke is ignored, an <Esc>
before an Alt+letter keystroke should be ignored too -- it should not
be interpreted as if the user had typed <Esc> <Esc> letter.
This fixes https://savannah.gnu.org/bugs/?54301.
Even when we don't act on most of these combinations, at least eat up
the whole sequence so the unknown keystroke doesn't enter things like
"3~" into the buffer.
This fixes https://savannah.gnu.org/bugs/?54291.
Also, add a period after the "Read nn lines" message, and
don't let the next shell prompt overwrite this message.
This addresses https://savannah.gnu.org/bugs/?53779.
When the keyboard buffer is empty, return a value that is not equivalent
to <Ctrl+Space>, so that typing a bunch of these while searching is going
on will not result in some of them getting executed afterward.
Instead of assuming that the entire contents of the keybuffer are an
escape sequence (even if this is far longer than any escape sequence
can be), remove from the keybuffer only those integers that actually
constituted a sequence, valid or invalid.
This fixes https://savannah.gnu.org/bugs/?48659.
There shouldn't be two separate variables (but with the same name) that
count the number of digits seen. Fuse them into a single static var.
This fixes https://savannah.gnu.org/bugs/?52863.
Each leading tab is converted to two tabs, and any leading four spaces
is converted to one tab. The intended tab size (for keeping most lines
within 80 columns) is now four.
When the mark is on, instead of letting a <Tab> simply insert a Tab
character at the cursor position, let it indent the marked region.
Original-idea-by: Chris Allegretta <chrisa@asty.org>
The pointer not being NULL is enough indication that the mark is set.
Also, rename the pointer from 'mark_begin' to simply 'mark', since
the former is kind of pleonastic.
Initialize 'breaking_col' to what it needs to be when the current chunk
consists of only a tab -- a tab that spreads across both start and end
of the chunk -- so that the last 'if' can move into the preceding one,
which allows the elision of 'char_len'.
Allow the user to record and run a single macro. The default binding
for starting and stopping the recording is M-: (Alt + colon) and for
running the macro M-; (Alt + semicolon).
This fulfills https://savannah.gnu.org/bugs/?50314.
Requested-by: Peter Passchier <peter@passchier.net>
Signed-off-by: Marco Diego Aurélio Mesquita <marcodiegomesquita@gmail.com>
Signed-off-by: Benno Schulenberg <bensberg@telfort.nl>
In atblanks mode, if a softwrapped chunk ends with a double-width space
character (say, Unicode 003000, Ideographic Space), and that would put
half of the character past the edge of the screen, break before it.
This fixes https://savannah.gnu.org/bugs/?51671.
The basic idea is that the cursor is always off, except when it needs
to be on: when waiting for text input, and in a few other cases: when
something was searched and found in the help viewer, and in the file
browser when option -g is in effect.
This fixes https://savannah.gnu.org/bugs/?51923.
Reported-by: Mike Frysinger <vapier@gentoo.org>
When multiple files were open and [x/n] was being shown in the title
bar, don't show nano's name and version number when just one buffer
remains open, but show [1/1] instead. It is less surprising.
When multiple buffers are open, replace nano's name and version number
with an indication how many buffers are open preceded by the sequence
number of the current buffer.
Signed-off-by: Marco Diego Aurélio Mesquita <marcodiegomesquita@gmail.com>
Signed-off-by: Benno Schulenberg <bensberg@telfort.nl>
Check the column number, not the character index, when suppressing the
line number on chunks other than the first, since the index can be zero
when in fact we're on a later chunk.
This fixes https://savannah.gnu.org/bugs/index.php?51790.
So that these functions have bindings by default, and easy bindings.
Add them to the help viewer too, so that searching backward becomes
possible there.
Tabs are of variable length depending on position, so allow them
to be displayed partially on one chunk and partially on another.
This fixes http://savannah.gnu.org/bugs/?51621.
Discarding (in commit 6f9bb53b) the cap on the number of chunks to
move backwards had as an unforeseen side effect that the screen can
fail to scroll when the cursor is somehow pushed offscreen.
Fix this by setting the target row (for smooth scrolling) always to
the bottom row of the edit window when nano notices that the cursor
has gone offscreen.
This fixes https://savannah.gnu.org/bugs/?51676.
Reported-by: David Lawrence Ramsey <pooka109@gmail.com>
The new option 'set selectedcolor' applies to marked text, to the
currently selected file in the file browser, and to the highlighted
match during interactive search-and-replace.
Commit 28beb3f added the 'forreal' parameter to prevent spotlight() from
placing the cursor wrongly due to an invalid placewewant. However, since
the variable-width softwrap overhaul (specifically, since commit 8490f4a),
place_the_cursor() no longer checks placewewant, so the parameter is no
longer needed.
Furthermore, dropping 'forreal' and thus always setting current_y won't
affect the operation of spotlight(), since the only functions that use
spotlight() (do_replace_loop() and do_int_spell_fix()) both call
edit_refresh() beforehand, which means that current_y will already
have been set to the value it will be set to again.
The function place_the_cursor() assumes that the viewport is up to date,
i.e., that current is in range of edittop. When uncutting or inserting,
however, place_the_cursor() gets called on the out-of-date viewport
first, and then a screen refresh is scheduled (which would put the
viewport up to date). This is backwards: the refresh should come before
the cursor placement, and the only reason it works anyway is because the
cap on the number of chunks to move backward papers over the problem by
keeping current_y in screen range regardless.
Fix this properly by simply setting current_y to the bottom row of the
screen instead of calling place_the_cursor(). This value of current_y
is only ever used when in smooth scrolling mode and the insertion (or
paste) pushed the cursor offscreen. In other situations, this value is
overridden when place_the_cursor() gets called after a screen refresh.
After that fix, the cap on the number of chunks to move backward is no
longer needed.
For some reason, when returning from suspension, SLang will produce
either a clipped error code (0xFF instead of 0xFFFF, when returning
from an externally induced suspension), or it will clip the code of
first subsequent keystroke to a single byte (when returning from a
normal, in-editor suspension: ^Z).
Side-step this by ignoring the clipped error code, and by using an
undefined control code as the first fake keystroke. Ignoring the
clipped error code is not possible when using a single-byte locale,
otherwise the user would not be able to type the character with
code 0xFF (although it could still be entered with Esc Esc 255).
This fixes https://savannah.gnu.org/bugs/?51477.
This should eliminate all cases where the newline is pushed off the edge
of the screen in non-atblanks softwrap mode. Also, it allows the use of
the last column of the screen in atblanks softwrap mode when a piece of
text does not contain any blanks.
This avoids the unreachable end-of-line reported on the mailing list:
http://lists.gnu.org/archive/html/nano-devel/2017-06/msg00011.html.
Extend get_softwrap_breakpoint() to break softwrapped lines on
whitespace when a flag is set. This flag is controlled by the new
rcfile option "atblanks". The '>' characters marking two-column
characters at the edge of the screen are disabled when it's on.
If get_softwrap_breakpoint() can't find whitespace in screen range, it
will break the line on the screen edge. (In this case, a blank can be
on the last column of the screen, but text can't, so that a blank on the
last column doesn't become invisible and possibly break the display.)
This fulfills https://savannah.gnu.org/bugs/index.php?49959.
Requested-by: Nicholas Boel <axxisd@gmail.com>
spotlight() now displays softwrapped lines chunk by chunk instead of all
at once. Since softwrapped lines are no longer of constant width, the
latter approach would fail if softwrapping breaks the spotlighted text.
Instead of taking a string, spotlight() now takes the starting and ending
columns of that string. Also, its handling of softwrapped lines is now
split off into a separate function, spotlight_softwrapped().
get_chunk_row() and get_chunk_leftedge() now become wrappers around
get_chunk(); the latter is only used directly in place_the_cursor()
when we need to get both the row and the leftedge. get_chunk() now
uses the proper formula to implement varying chunk width.
Since chunk width now varies, place_the_cursor() sets the x position
relative to the leftedge in a different way that works regardless
of chunk width, update_softwrapped_line() loops until it gets all
softwrap breakpoints instead of calculating the full length in
advance and getting one full row at a time, the chunk iterators
now count leftedges instead of rows, and fix_firstcolumn() does a
full recalculation of the chunk that firstcolumn is on instead of
simply shifting it back.
Also, in update_softwrapped_line(), when a line's softwrap breakpoint
is before the last column of the edit window, a ">" is now added to
the end of it.
The workaround in place_the_cursor() for when two-column characters
straddle the edge of the screen is removed, as it's no longer needed
now that chunks end before such characters.
Furthermore, do_home() and do_end() use xplustabs() instead of
placewewant again when calculating the leftedge, since placewewant
refers to a column that may or may not be available (if it's not,
the cursor will be placed wrongly). Make get_edge_and_target() use
xplustabs() instead of placewewant for the same reason; this also lets
us simplify get_edge_and_target(), since xplustabs() will never be
greater than strlenpt(). Finally, since do_end() now has to calculate
rightedge as well as rightedge_x, use the former to implement the same
time-saving optimizations as in do_home().
The cursor is not yet adjusted when we try to go directly to a column
past the end of a softwrap breakpoint, and placewewant handling in the
vertical movement code is not yet adjusted for varying chunk lengths,
but fixes for these are forthcoming.
This fixes https://savannah.gnu.org/bugs/?49440.
get_chunk_row() replaces the formula "column / editwincols".
get_chunk_leftedge() replaces "(column / editwincols) * editwincols".
get_last_chunk_row() replaces "strlenpt() / editwincols".
get_last_chunk_leftedge() replaces "(strlenpt() / editwincols) * editwincols".
This prepares us for any changes in those formulas, and for more such
functions later.
The new function find_softwrap_breakpoint() returns the column number
of the last position in screen range where we can wrap the given text
without breaking a two-column character in half (as was done until now).
The returned column number is the leftedge of the next softwrapped chunk.
If the end of the text is reached while searching for a wrapping point,
the parameter end_of_line is set to TRUE.
The new function get_chunk() uses find_softwrap_breakpoint() to find the
row and leftedge corresponding to a given column of a given line.
When also --nohelp is active and the terminal is so narrow that the
message that --constantshow displays on the bottom row does not fit,
this causes the cursor to be pushed "offscreen". Some terminal
emulators don't handle this case correctly, and leave the cursor
in an invisible or mistaken position. Compensate for this by
moving the cursor back to the start of the row.
This fixes https://savannah.gnu.org/bugs/?51335.
Reported-by: David Lawrence Ramsey <pooka109@gmail.com>
When receiving a SIGCONT, don't call regenerate_screen() directly
but act as if a SIGWINCH occurred. Also insert a dummy key, and
thus let the relevant input routine redraw the relevant subwindows.
Use KEY_F0 as the dummy key, as both Ncurses and Slang know it, and
it should normally do nothing at all.
This fixes https://savannah.gnu.org/bugs/?51124.
When we are in the file browser, browser_refresh() will take care of
redrawing the title bar and the file list, so total_refresh() should
leave those two subwindows alone in that case.
This fixes https://savannah.gnu.org/bugs/?51147.
Achieve this by making the suppression flag global, so that we can
just reset it instead of making an improper call of do_cursorpos().
This fixes the secondary part of https://savannah.gnu.org/bugs/?51134.
When spotlighting the string to be replaced, placewewant isn't valid,
so tell place_the_cursor() to ignore its value to avoid the cursor
getting mistakenly placed at the beginning of the next row.
This fixes https://savannah.gnu.org/bugs/?50997.
Reported-by: David Lawrence Ramsey <pooka109@gmail.com>
When the editing space gets narrower, either by toggling line numbers on
or making nano's window smaller, and when the cursor is near the bottom,
it can be pushed offscreen. In that case, don't get it back onscreen by
centering the current line but by putting it on the bottom row. This
makes for a smoother experience.
This fixes http://savannah.gnu.org/bugs/?50933.
This avoids https://savannah.gnu.org/bugs/?49912 while at the same time
avoiding to draw the edit window twice in a row -- the first drawing
would use a wrong margin, which results in a visible and irritating
shift left or right of the content upon the second drawing.
This fixes https://savannah.gnu.org/bugs/?50877.
An iterator should not be called "start_col", because it is only the
starting column at the very beginning.
Also, start_col (after the rename) can never be /larger/ than column.
Allow the user to search in a help text with ^W and M-W.
Achieve this by not writing the help text directly to the screen
but first writing it to a temporary file and then opening this file
in a new buffer, and treating it specially: the normal file-reading
feedback is suppressed, the titlebar shows the headline of the text,
the cursor is hidden, and the menu is limited to just the up and down
movements and searching.
This fulfills https://savannah.gnu.org/bugs/?28994.
Signed-off-by: Rishabh Dave <rishabhddave@gmail.com>
To avoid an ncurses hiccup (miscoloring) when running on musl.
This fixes https://savannah.gnu.org/bugs/?50787.
Reported-by: Avi Halachmi <avihpit@yahoo.com>
The cursor needs to be placed in its proper spot in the edit window
/only/ when nano is about to accept input from the user and needs to
show where this input will go.
(This might cause a scrolling issue to appear, because reset_cursor()
does not just place the cursor, it also recomputes current_y, which
is used in several places to determine whether and how much to scroll.
If it so happens, we'll deal with that fallout later.)
The interval 2013-2017 for the Free Software Foundation is valid
because in those years there were releases with changes by either
Chris or David, and the GNU maintainers guide advises to mention
a new year in all files of a package, not just in the ones that
actually changed, and be done with it for the rest of the year.
On some terminal emulators, Ctrl+Home and Ctrl+End produce special
keycodes, distinct from plain Home and End. Make the users of those
emulators (and of the Linux console) glad by making ^Home and ^End
do the obvious thing, and the combinations with Shift too.
When a multi-column character straddles a chunk boundary, and the
preferred column (placewewant) for the cursor is zero, cheat: show
the cursor not where the character starts but on the beginning of
the next row. This makes the cursor move smoothly in the leftmost
column of the screen when using <Up> and <Down> and such, instead
of jumping around.
In this way the scrolling logic won't get confused and the screen
will scroll properly when stepping beyond the top or bottom row.
This fixes https://savannah.gnu.org/bugs/?50687.
If edittop is partially offscreen before we scroll, and it gets
scrolled more offscreen, we do need to compensate for the chunks
between firstcolumn and leftedge -- that is: the chunks between
the top row and the cursor row.
This fixes https://savannah.gnu.org/bugs/?50691.
When a two-column character cannot be shown because it straddles the
boundary between two chunks of a line, show the '>' placeholder for
its left "half", and '<' for its right "half".
This mitigates https://savannah.gnu.org/bugs/?49440.
The number of rows to draw shouldn't be compensated for the chunks
of edittop that are before firstcolumn, because they are offscreen.
This completes the fix for https://savannah.gnu.org/bugs/?50621.
There is no need to always increase nrows by 1 or 2 -- an increase
of 1 is only needed when the line that borders on the scrolled region
needs to redrawn too: when this line was horizontally scrolled or when
the mark is on.
This fixes https://savannah.gnu.org/bugs/?50621.
Reported-by: David Lawrence Ramsey <pooka109@gmail.com>
If the number of columns in the edit window changes (which currently
only happens in two places: in regenerate_screen(), called when the
window is resized; and in main(), when line numbering mode is toggled),
the display will break if we're in softwrap mode and firstcolumn is
nonzero. This is because the column width of softwrapped chunks has
changed, and firstcolumn is no longer the starting column of a chunk,
an assumption that all code using firstcolumn relies on.
To fix this problem, add a new function, ensure_firstcolumn_is_aligned(),
to adjust firstcolumn to the starting column of the chunk it's on, and
use it when the number of columns in the edit window changes.
(Note that this function uses the simplest possible fix, and could
probably be made more sophisticated.)
The new function, update_softwrapped_line(), is called from inside
update_line() when softwrap mode is on, so that existing calls remain
unchanged. It takes no index, instead displaying edittop from column
firstcolumn, and all other lines from column zero.
If current is on edittop, it's displayed using the edittop rules, but
this is not a problem: if current[current_x] is above edittop at column
firstcolumn, it's offscreen, and that should be handled before calling
update_line() anyway.
Together with the preceding bunch of changes,
this fixes https://savannah.gnu.org/bugs/?47667.
Make current_is_above_screen() check for current[current_x] being above
edittop at column firstcolumn, and make current_is_below_screen() start
counting down from edittop at column firstcolumn instead of edittop at
column zero. This means that both functions now account for softwrapped
chunks properly.
Actually enable scrolling edittop partially off the screen by making
edit_scroll() and adjust_viewport() use firstcolumn properly when
iterating through softwrapped chunks in softwrap mode, or lines in
non-softwrap mode.
In non-softwrap mode, firstcolumn should still always be zero, because
it's initially set to that, and because passing it through the iterators
will maintain it at that.
This fixes https://savannah.gnu.org/bugs/?49100.
Reported-by: David Lawrence Ramsey <pooka109@gmail.com>
Since all lines can be partially scrolled off the screen now
(except for edittop, which is forthcoming), the maxlines global
variable and its computation mechanism are no longer needed.
Use go_back_chunks() to adjust edittop, instead of special casing
the computation of goal when softwrapping. Now softwrap mode and
non-softwrap mode will behave the same way when edittop can be
partially scrolled off the screen, which is forthcoming.
(Note that the top line of the screen can't be partially scrolled
yet, so we have to work around that for now.)
Use go_back_chunks() and go_forward_chunks() to adjust edittop and to
move up or down to the scrolled region before updating the rows there.
Now softwrap mode and non-softwrap mode will behave the same way when
we can scroll the top line of the screen partially off the screen,
which is forthcoming.
(Note that the top line of the screen can't be partially scrolled yet,
so we have to work around that for now.)
Not drawing a line on a row if we're on the top row and scrolled down,
or if we're on the bottom row and scrolled up, will only work properly
if the line on that row takes up only that row. The latter might not
be the case in softwrap mode: if the line occupies multiple chunks and
begins on that row -- in that case none of the chunks would be drawn.
Add the new functions current_is_above_screen() (which doesn't account
for softwrapped chunks yet, but will when we can scroll edittop partially
off the screen, which is forthcoming), current_is_below_screen() (which
determines whether current[current_x] is past the softwrapped chunk at
the bottom of the screen), and current_is_offscreen() (the union of the
previous two functions).
edit_redraw() and edit_refresh() now use current_is_offscreen() to check
whether they should adjust the viewport, and adjust_viewport() now uses
current_is_above_screen() to determine whether current is on or below
the screen in FLOWING mode.
Add the new function less_than_a_screenful() to accomplish this.
It uses go_back_chunks() to count the number of softwrapped chunks
between the end point and the starting point of the paste.
Now softwrap mode and non-softwrap mode behave the same way when
uncutting fewer than editwinrows rows of text. Accordingly, remove
the call to ensure_line_is_visible(), as it no longer applies.
These functions, go_back_chunks() and go_forward_chunks(), take a number
of softwrapped chunks (screen rows) to move, a pointer to a buffer, and
a location (specifically, a starting column of a softwrapped chunk). If
they move successfully, they will update the buffer pointer and location
to point to the beginning of the softwrapped chunk they moved to.
Since non-softwrap mode is effectively just a subset of softwrap mode
in which every line takes up one chunk, these functions also work in
non-softwrap mode. In this case, their starting column will always be
zero, as it would be in softwrap mode on a line that takes up one chunk.
Nothing uses these functions yet, but that is forthcoming.
Instead of redetermining the entire span of the converted string,
simply move one character left, and then bite it off to make place
for the trailing $.
If the last two columns of a row would be taken up by a double-width
character (and the line is longer than that), don't print it, because
it wouldn't leave any room for the $ character.
This fixes https://savannah.gnu.org/bugs/?50491.
Reported-by: David Lawrence Ramsey <pooka109@gmail.com>
Running strlenpt() on a string that takes up more than 80 columns
(the width of an average terminal) takes /more/ time than simply
converting an extra character (the one that will be overwritten
by the "$" at the edge of the terminal). So... just convert one
more character than necessary when the line is overlong. In the
most common case, however, the line will fit fully onscreen, and
we save a whole call of strlenpt().
Since we only need span columns of the string, stop scanning the string
as soon as we have that many columns, instead of scanning the string all
the way to the end. This speeds up the conversion of very long lines.
Benno Schulenberg