…::AbsoluteLine`

An `AbsoluteLine` represents an index (or length) into the complete scrollback buffer, rather than just the so-called 'active region'. This will be used in a select few methods within `Grid<T>`, while the rest of them are oblivious to whether scrollback is enabled or not, and continue to use `Line`.

Given this distinction, I wonder if one should go further and distinguish between a `Line` as an index, and `NumLines` as a length. But perhaps that's too much verbosity.

That said, idk why exactly it was unsafe to start with so perhaps this is incorrect.

It hasn't been published to crates.io yet.

This is accomplished by replacing the `Vec` buffer with a `VecDeque`.

The *active region* represents the region in the buffer that is modifiable. It is essentially the latest `self.num_lines()` lines of the buffer.

Upon start up, the *visible region* is equal to the active region, and when new lines are added the visible/active region both move accordingly. However when you scroll, the visible region changes. When you type in new commands again tho, the visible region jumps back to equal the active region.

This might have messed up the 'selected' colours... Have to check later

Now they work properly with `lines` > 1.

This completely ignores existing mouse input handling (eg: the `normal_report` function or `scroll_px` field) and only works with a `LineDelta`. Additionally, there is a hardcoded scroll sensitivity of 5.0. This should probably go in the config.

Already I've noticed there is flickering every now and then while scrolling. I'll have to investigate what is causing this.

Now they compile.. Running: they all fail atm :D

It incorrectly reported `AtBottom` or `AtTop`, meaning that the dirty flag was not set and the terminal wasn't re-rendered.

Now the scroll logic is split into three cases: (a) its already at the top/bottom (b) after this movement it *will* be at the top/bottom (c) none of the above.

thanks jwilm, I tried make the implementation a little more DRY, hopefully I didn't break anything through doing so.

This makes it easier to understand what is going on. Instead of a distance from the end to the end of the visible region (scroll_back_amount), we simply store the starting cell of the visible region (visible_region_start). This makes various calculations easier.

Now it jumps to the bottom when pressing escape.

Also some unecessary variables were removed.

Now the window resize algorithm has been built around what I've observed in Gnome Terminal (and I imagine other terminals too)... The algorithm tries to make the bottom line stay the same when growing/shrinking the view. Sometimes this means the cursor needs to be clipped.

There's one issue at the moment though: if you open a new terminal, enter fullscreen, then exit fullscreen - there will be an extra prompt in the scrollback buffer. This is really only a cosmetic issue, but still something I'd like to fix down the track. The only way I've thought of fixing this would be to aggressively delete those empty lines when exiting from fullscreen. Perhaps someone else can think of something?

Also modified the default key bindings to use those and `ScrollUp`/`ScrolDown`... The problem is, this overrides some of the existing keybindings. Which ones should take precedence?

The rest of the failing tests happen when the test case causes
scrolling. Not sure how to modify Grid to fix those.

At the moment, enabled=false is somewhat buggy. ie: it still will scroll if the window is resized...

…ed/swapped

This is needed for fixing selections.

Sneak peek: the new selections system will use `AbsolutePoint` instead of `Point` and so you'll be able to select anywhere in the grid, even outside the visible/active regions.

Now all selections operate on `AbsoluteLine`s

…::AbsoluteLine`

An `AbsoluteLine` represents an index (or length) into the complete scrollback buffer, rather than just the so-called 'active region'. This will be used in a select few methods within `Grid<T>`, while the rest of them are oblivious to whether scrollback is enabled or not, and continue to use `Line`.

Given this distinction, I wonder if one should go further and distinguish between a `Line` as an index, and `NumLines` as a length. But perhaps that's too much verbosity.

That said, idk why exactly it was unsafe to start with so perhaps this is incorrect.

It hasn't been published to crates.io yet.

This is accomplished by replacing the `Vec` buffer with a `VecDeque`.

The *active region* represents the region in the buffer that is modifiable. It is essentially the latest `self.num_lines()` lines of the buffer.

Upon start up, the *visible region* is equal to the active region, and when new lines are added the visible/active region both move accordingly. However when you scroll, the visible region changes. When you type in new commands again tho, the visible region jumps back to equal the active region.

This might have messed up the 'selected' colours... Have to check later

Now they work properly with `lines` > 1.

This completely ignores existing mouse input handling (eg: the `normal_report` function or `scroll_px` field) and only works with a `LineDelta`. Additionally, there is a hardcoded scroll sensitivity of 5.0. This should probably go in the config.

Already I've noticed there is flickering every now and then while scrolling. I'll have to investigate what is causing this.

Now they compile.. Running: they all fail atm :D

It incorrectly reported `AtBottom` or `AtTop`, meaning that the dirty flag was not set and the terminal wasn't re-rendered.

Now the scroll logic is split into three cases: (a) its already at the top/bottom (b) after this movement it *will* be at the top/bottom (c) none of the above.

thanks jwilm, I tried make the implementation a little more DRY, hopefully I didn't break anything through doing so.

This makes it easier to understand what is going on. Instead of a distance from the end to the end of the visible region (scroll_back_amount), we simply store the starting cell of the visible region (visible_region_start). This makes various calculations easier.

Now it jumps to the bottom when pressing escape.

Also some unecessary variables were removed.

Now the window resize algorithm has been built around what I've observed in Gnome Terminal (and I imagine other terminals too)... The algorithm tries to make the bottom line stay the same when growing/shrinking the view. Sometimes this means the cursor needs to be clipped.

There's one issue at the moment though: if you open a new terminal, enter fullscreen, then exit fullscreen - there will be an extra prompt in the scrollback buffer. This is really only a cosmetic issue, but still something I'd like to fix down the track. The only way I've thought of fixing this would be to aggressively delete those empty lines when exiting from fullscreen. Perhaps someone else can think of something?

Also modified the default key bindings to use those and `ScrollUp`/`ScrolDown`... The problem is, this overrides some of the existing keybindings. Which ones should take precedence?

The rest of the failing tests happen when the test case causes
scrolling. Not sure how to modify Grid to fix those.

At the moment, enabled=false is somewhat buggy. ie: it still will scroll if the window is resized...

…ed/swapped

This is needed for fixing selections.

Sneak peek: the new selections system will use `AbsolutePoint` instead of `Point` and so you'll be able to select anywhere in the grid, even outside the visible/active regions.

Now all selections operate on `AbsoluteLine`s

Rebased neon64/alacritty on top of the latest alacritty master. These
are just some minor tweaks I had to do manually after merging.

Ref tests were broken because it assumed for existing tests that a
scrollback buffer is present. To fix this a check for tests without
scrollback has been added. Once a ref without scrollback is detected,
only the visible buffer is compared.

…::AbsoluteLine`

An `AbsoluteLine` represents an index (or length) into the complete scrollback buffer, rather than just the so-called 'active region'. This will be used in a select few methods within `Grid<T>`, while the rest of them are oblivious to whether scrollback is enabled or not, and continue to use `Line`.

Given this distinction, I wonder if one should go further and distinguish between a `Line` as an index, and `NumLines` as a length. But perhaps that's too much verbosity.

That said, idk why exactly it was unsafe to start with so perhaps this is incorrect.

It hasn't been published to crates.io yet.

This is accomplished by replacing the `Vec` buffer with a `VecDeque`.

The *active region* represents the region in the buffer that is modifiable. It is essentially the latest `self.num_lines()` lines of the buffer.

Upon start up, the *visible region* is equal to the active region, and when new lines are added the visible/active region both move accordingly. However when you scroll, the visible region changes. When you type in new commands again tho, the visible region jumps back to equal the active region.

This might have messed up the 'selected' colours... Have to check later

Now they work properly with `lines` > 1.

This completely ignores existing mouse input handling (eg: the `normal_report` function or `scroll_px` field) and only works with a `LineDelta`. Additionally, there is a hardcoded scroll sensitivity of 5.0. This should probably go in the config.

Already I've noticed there is flickering every now and then while scrolling. I'll have to investigate what is causing this.

Now they compile.. Running: they all fail atm :D

It incorrectly reported `AtBottom` or `AtTop`, meaning that the dirty flag was not set and the terminal wasn't re-rendered.

Now the scroll logic is split into three cases: (a) its already at the top/bottom (b) after this movement it *will* be at the top/bottom (c) none of the above.

thanks jwilm, I tried make the implementation a little more DRY, hopefully I didn't break anything through doing so.

This makes it easier to understand what is going on. Instead of a distance from the end to the end of the visible region (scroll_back_amount), we simply store the starting cell of the visible region (visible_region_start). This makes various calculations easier.

Now it jumps to the bottom when pressing escape.

Also some unecessary variables were removed.

Now the window resize algorithm has been built around what I've observed in Gnome Terminal (and I imagine other terminals too)... The algorithm tries to make the bottom line stay the same when growing/shrinking the view. Sometimes this means the cursor needs to be clipped.

There's one issue at the moment though: if you open a new terminal, enter fullscreen, then exit fullscreen - there will be an extra prompt in the scrollback buffer. This is really only a cosmetic issue, but still something I'd like to fix down the track. The only way I've thought of fixing this would be to aggressively delete those empty lines when exiting from fullscreen. Perhaps someone else can think of something?

Also modified the default key bindings to use those and `ScrollUp`/`ScrolDown`... The problem is, this overrides some of the existing keybindings. Which ones should take precedence?

The rest of the failing tests happen when the test case causes
scrolling. Not sure how to modify Grid to fix those.

At the moment, enabled=false is somewhat buggy. ie: it still will scroll if the window is resized...

…ed/swapped

This is needed for fixing selections.

Sneak peek: the new selections system will use `AbsolutePoint` instead of `Point` and so you'll be able to select anywhere in the grid, even outside the visible/active regions.

Now all selections operate on `AbsoluteLine`s

In the xterm specification the escape sequence 1049 is used to switch to
an alternate buffer and back. This is used by applications like `man` or
`less` for hijacking the scrollback.

With this commit whenever 1049h is sent, scrollback is disabled. And
sending 1049l will then enable scrollback again and switch back to the
normal buffer.

Seems like the current Cargo.lock file doesn't work. So here's the
updated version that should make everything work properly.

Also only resets it to true if scrollback was enabled in the first place

Added some additional tests. Most of these tests check the same thing as
the existing tests but take scrollback into account.

The vim_scrollback_scrolling_disabled check makes sure the normal
scrollback is disabled inside of vim when the alternate screen buffer is
active.

The alternate_screen_buffer_(enter|exit) make sure the grid changes to
the alternate buffer and back when the 1049 escape sequence is sent.

Add additional scrollback tests

Currently scrollback tests were failing because some fields were missing
and the `currently_enabled` flag was mismatched.
This is a first attempt at fixing this, with this commit all test work
again.

The detection for old ref tests now works properly without any false
positives. All tests still work.

I'm not 100% certain why this test was failing, but after redoing the
test with a little different approach, everything seems to be going
right now.

All existing ref-test have been manually edited so they now have the
scrollback field. This should be a better way to test the old tests with
scrollback.

Fix failing ref tests

This is still a leftover from testing the ref test changes.

Issues with resize:
`self.total_lines_in_buffer` was being used to calculate whether new lines should be added. This meant that lines weren't added when they needed to be.

`move_visible_region_up` was actually used internally by `grow_lines`, and @chrisduerr's PR accidentally disabled it when scrollback was disabled. So I've moved the scrollback enabled check to inside `Term` instead. Idk if this is good in the future or not.

Renames: `total_lines_in_buffer()` -> `num_absolute_lines()`, since not all those absolute lines were actually being stored in the buffer (it was misleading).

`min_line` -> `min_absolute_line` -> reflects what it actually returns.. the minimum line in the active/visible region will always be `Line(0)` - we don't need a function for that.

Also added an assertion that: `self.raw.len() >= self.absolute_to_raw_index(self.visible_region().end)`

Also got rid of an extra `impl<T> Grid<T>` block, by moving its contents into the other one.

…k enabled check

This is still a leftover from testing the ref test changes.

The ref tests were failing because the `get_scrollback_enabled` method
was checking for currently enabled scrollback, not absolutely enabled
scrollback. This has been fixed by introducing a separate method for
checking if it's enabled.

To prevent this issue in the future the `get_scrollback_enabled` has
also been renamed to `scrollback_currently_enabled` to make clear what
is actually being tested.

There were also a few public statements sprinkled around that were not
necessary, those have been removed.

…ollback

Fix failing ref tests