GitLab

Since commit 41ed6907, cancelling a prompt after tabbing would sometimes
leave the list of file names on the screen.  When testing this first,
it worked fine -- I was fooled again by 'edit_refresh_needed' already
being TRUE when nano has just started up and sits waiting for the first
key stroke.  I have to hunt this down and kill it.

Since commit dac0aa1d, nano would leave edittop at current after reading
data from standard input, meaning that everything that had been read in
was invisible, "scrolled off" the top of the screen.  Correct this by
explicitly setting edittop.

The above does not happen for ^R (^X) because it has a special mechanism
to save and restore edittop.  Nor does it happen for ^R ^X M-F because it
sets current to the top of the file and refreshing the screen will make
edittop follow, or -- in case just one unterminated line was read in --
edittop will already be equal to current.

Names could be strings of multibyte characters, so they could be more
than COLS - 1 bytes long and still be less than COLS - 1 columns wide.

Also unwrap or improve some comments.

A normal lock file is apparently 1024 bytes in size, so the second
attempt at reading bytes from the file would try to read 8192 more
bytes into a buffer that has room for only 7168 left.  According to
valgrind, the read() function doesn't like that -- and true: if for
some reason the lock file had suddenly expanded, the buffer would
overflow.

This fixes https://savannah.gnu.org/bugs/?47156.

When a tilde is used in the name given at the ^R or ^O prompts, nano
expands it, but /not/ when a tilde is given on the command line (in
such a way that the shell leaves it as is).  Correct that asymmetry.

This fixes https://savannah.gnu.org/bugs/?44929
and fixes https://savannah.gnu.org/bugs/?47702
and fixes https://savannah.gnu.org/bugs/?47771.

Commit 36ec76a5 made the wrong change: after a tab that did not list any
file names on the screen, a refresh /is/ needed, because a previous tab
might have listed things on the screen.  But at the end of the prompt,
it is not necessary to refresh the edit window if things were listed,
because the window will be refreshed anyway after reading in a file.

Use 'slash' to point at a possible slash, use 'filename' just to
point at the real file name, and use 'wasdirname' just to point at
the dir's name before expanding it in order to be able to free it.

Also, remove two superfluous asserts: 'dirname' cannot be NULL
because it has just been mallocstrcpy'd, and checking 'num_matches'
is pointless as it would crash on the next statement anyway.

This is a remnant from 2001, when things were different.  Now, there
is no need to refresh the edit window when tabbing produced no list.
When it did produce a list, it is cleared off later.

If for some reason opening the spell-checked or formatted file fails,
don't throw away the current contents of the buffer.

(It should also give proper feedback about the failure, but we'll leave
that for some other time.)

Because it is a little clearer, and it is what Pico does too.

This partly fixes https://savannah.gnu.org/bugs/?47721.

This fixes https://savannah.gnu.org/bugs/?47720.

Also, store the input character earlier, so we don't have to use len - 1.
Furthermore, len increments in steps of 1, so it cannot pass the value of
bufx unnoticed, so use a comparison for equality.

Just let read_line() zero-terminate the intermediate buffer when
the line is complete.

Most of the time NO_CONVERT will not be set, the number of lines will
not be zero, and the format of the file will be zero.  Rearrange the
conditions so that they will evaluate as FALSE as soon as possible.

Index i follows almost synchronously the value of len.  Since we're
adding characters to the intermediate buffer always only at the end,
just use len as the index.

Until now (when not leaving files unconverted), nano would fumble and
drop the final carriage return of a Mac file, and would thus treat the
last line of such a file as an unterminated line and prepend it to the
current line of the buffer.  Correct that, and delete the dead piece
of code that was meant to do this.

This fixes https://savannah.gnu.org/bugs/?47716.

When we don't set edittop in read_line(), we don't need to readjust it in
read_file(), because in that particular case it will still be pointing at
current.  And since fileptr is a new, freshly created line, it can never
be equal to filebot, so there is no point in comparing them.

If more than one line was inserted at the beginning of the file, leave it
up to the screen handling to set edittop to what it should be.

Move the setting of fileage a bit down, to its sister setting: the line
at current gets "connected" either to the top-of-file pointer or to the
last line of the inserted file.

Also don't zero-terminate the matches in order to compare them,
but just limit the length of the comparison.

The number of lines to scroll is: the y position of the start of the
current line, plus the extra lines that this line occupies, plus the
extra lines that the next line occupies, plus one, minus the y position
of the last window line.

The y position of the start of the current line is current_y -
xplustabs() / COLS, the extra lines are strlenpt(data) / COLS,
and the y position of the last window line is editwinrows - 1.

Note that we first compute the amount to scroll before actually moving
to the next line, because we need the current value of current_x, not
the one that it will have in the next line.  The placewewant value is
not good either, because it might be beyond where we actually are.

This fixes https://savannah.gnu.org/bugs/?47665.

(This change will be made superfluous when we start using gnulib.)

This prevents getcwd() from failing on Android and thus completes the
fix for https://savannah.gnu.org/bugs/index.php?47659

.
Reported-by: Chris Renshaw <osm0sis@outlook.com>
Signed-off-by: Benno Schulenberg <bensberg@justemail.net>

Doing a chdir("..") will not fail when the root directory is reached,
and when getcwd() keeps failing too, we have no way of knowing when
to stop.  So, simply limit the number of attempted chdirs, to avoid
getting into an endless loop.

This avoids the hang in https://savannah.gnu.org/bugs/index.php?47659

.
Reported-by: Chris Renshaw <osm0sis@outlook.com>
Signed-off-by: Benno Schulenberg <bensberg@justemail.net>

If nano has less than four columns available, it will die,
so there will always be room for at least four characters.

Also, don't force a full refresh of the edit window simply because the
current line needs to be horizontally scrolled.  And further, when the
adjustment of edittop has determined that a full refresh is needed,
get out and don't bother scrolling some lines first.

When in softwrap mode and scrolling down a line, and thus going to
do a full refresh, get out and don't bother redrawing the current
and prior lines first.

When moving the cursor up or down one line, redraw the new current
line only when the target column (placewewant) is beyond the screen,
or when the mark is on.

(This still redraws the current and prior lines unnecessarily when
they are in fact shorter than the screen is wide and the mark is off,
but we'll let that pass for now.)

Also, when softwrap is on, we don't have have to redraw the current
and prior lines at all (when the mark is off): they are in full view,
there is nothing to show or hide.

When scrolling down a line, a full refresh of the edit window is only
needed when softwrap is on, because only then the movement is irregular.
When each file line takes up just one screen line (softwrap is off),
edit_scroll() is perfectly able to scroll and redraw only the necessary
lines.

(But... when doing a full refresh anyway with softwrap, why bother
scrolling at all?  Why not just adjust edittop and call refresh?)

These are formats used by binutils/glibc/gdb/gcc.

The old_current line needs to be redrawn only if it differs from current,
and if it wasn't drawn already by the iteration for when the mark is on.

Also make the conditions involving horizontal scrolling more precise.

Instead of saving the current value of placewewant, then setting the
new value, and then passing the old value to edit_redraw() in seven
different places, just let edit_redraw() do this saving and setting.

In the bargain placewewant is now only recalculated when it matters
-- when allow_update is TRUE -- and not when it's superfluous.