notcurses/CHANGELOG.md
nick black f63325db60 Lots of ncvisual work
Unify ffmpeg/oiio/null implementations, where possible. This effectively
required placing all three in the same file, which meant they're all now
C++. Update FFmpeg implemenation to be C++-usable. Implement
ncvisual_rotate_cw() and ncvisual_rotate_ccw() #515. Move most of tetris
over to Visual from Plane #558. Add bgra_to_rgba(), necessary for
creating ncvisual from BGRA memory. Implement ncvisual_from_rgba()
and ncvisual_from_bgra() #557. Add unit tests on ncvisual rotation.
2020-05-06 01:02:46 -04:00

6.3 KiB

This document attempts to list user-visible changes and any major internal rearrangements of Notcurses.

  • 1.3.4 (not yet released)

    • notcurses_lex_margins() has been added to lex margins expressed in either of two canonical formats. Hopefully this will lead to more programs supporting margins.
    • ncvisual_open_plane() has been renamed ncvisual_from_file(). The former has been retained as a deprecated alias. It will be removed by 1.6/2.0.
    • ncvisual_from_rgba() and ncvisual_from_bgra() have been added to support creation of ncvisuals from memory, requiring no file.
    • ncvisual_rotate_cw() and ncvisual_rotate_ccw() have been added, having the same semantics as ncplane_rotate_cw() and ncplane_rotate_ccw(). They will likely be augmented in the future to support arbitrary rotations.
    • ncvisual_from_plane() has been added to support "promotion" of an ncplane to an ncvisual. The source plane may contain only spaces, half blocks, and full blocks.
  • 1.3.3 (2020-04-26)

    • The ncdplot type has been added for plots based on doubles rather than uint64_ts. The ncplot type and all ncplot_* functions were renamed ncuplot for symmetry.
    • FFMpeg types are no longer leaked through the Notcurses API. AVERROR is no longer applicable, and ncvisual_decode() no longer returns a struct AVframe*. Instead, the nc_err_e enumeration has been introduced. Functions which once accepted a value-result AVERROR now accept a value- result nc_err_e. The relevant constants can be found in notcurses/ncerrs.h.
    • OpenImageIO 2.1+ is now supported as an experimental multimedia backend. FFmpeg remains recommended. Video support with OIIO is spotty thus far.
    • CMake no longer uses the USE_FFMPEG option. Instead, the USE_MULTIMEDIA option can be defined as ffmpeg, oiio, or none. In cmake-gui, this item will now appear as an option selector. oiio selects OpenImageIO.
  • 1.3.2 (2020-04-19)

    • ncdirect_cursor_push(), notcurses_cursor_pop(), and ncdirect_cursor_yx() have been added. These are not supported on all terminals. ncdirect_cursor_yx() ought be considered experimental; it must read a response from the terminal, and this can interact poorly with other uses of standard input.
    • 1.3.1 unintentionally inverted the C++ Notcurses::render() wrapper's return code. The previous semantics have been restored.
  • 1.3.1 (2020-04-18)

    • ncplane_at_yx() and ncplane_at_cursor() have been changed to return a heap-allocated EGC, and write the attributes and channels to value-result uint32_t* and uint64_t* parameters, instead of to a cell*. This matches notcurses_at_yx(), and means they're no longer invalidated if the plane in question is destroyed. The previous functionality is available as new functions ncplane_at_yx_cell() and ncplane_at_cursor_cell().
    • ncplane_set_base() inverted its uint32_t attrword and uint64_t channels parameters, thus matching every other function with these two parameters. It moved const char* egc before either, to force a type error, as the change would otherwise be likely to go overlooked.
    • Scrolling is now completely implemented. When a plane has scrolling enabled through use of ncplane_set_scrolling(true), output past the end of the last line will now result in the top line of the plane being lost, all other lines moved up one, and the bottom line cleared.
  • 1.3.0 (2020-04-12)

    • No user-visible changes
  • 1.2.9 (2020-04-11)

    • No user-visible changes
  • 1.2.8 (2020-04-10)

    • notcurses-tetris now happily continues if it can't load its background.
  • 1.2.7 (2020-04-10)

    • Plots now always keep the most recent data to their far right (i.e., the gap that is initially filled is on the left, rather than the right).
  • 1.2.6 (2020-04-08)

    • ncplane_putsimple_yx() and ncplane_putstr_yx() have been exported as static inline functions.
    • ncplane_set_scrolling() has been added, allowing control over whether a plane scrolls. All planes, including the standard plane, do not scroll by default. If scrolling is enabled, text output via the *put* family of functions continues onto the next line when encountering the end of a row. This does not apply to e.g. boxes or lines.
    • ncplane_putstr_yx() now always returns the inverse of the number of columns advanced on an error (it used to return the positive short count so long as the error was due to plane geometry, not bad input).
    • ncplot_add_sample() and ncplot_set_sample() have been changed to accept a uint64_t rather than int64_t, since negative samples do not currently make sense. Plots were made more accurate in general.
    • notcurses_term_dim_yx() now accepts a const struct notcurses*.
    • notcurses_resize() is no longer exported. It was never necessary to call this in response to a resize, despite confusing documentation that could have been read to suggest otherwise. If you're in a long block on input, and get an NCKEY_RESIZE, just call notcurses_refresh() (which now calls notcurses_resize() internally, as notcurses_render() always has).
    • First Fedora packaging.
  • 1.2.5 (2020-04-05)

    • Add ncplot, with support for sliding-windowed horizontal histograms.
    • gradient, polyfill, ncplane_format() and ncplane_stain() all now return the number of cells written on success. Failure still sees -1 returned.
    • ncvisual_render() now returns the number of cells emitted on success, as opposed to 0. Failure still sees -1 returned.
    • ncvisual_render() now interprets length parameters of -1 to mean "to the end along this axis", and no longer interprets 0 to mean this. 0 now means "a length of 0", resulting in a zero-area rendering.
    • notcurses_at_yx() no longer accepts a cell* as its last parameter. Instead, it accepts a uint32_t* and a uint64_t*, and writes the attribute and channels to these parameters. This was done because the gcluster field of the cell* was always set to 0, which was surprising and a source of blunders. The EGC is returned via the char* return value. https://github.com/dankamongmen/notcurses/issues/410
  • 1.2.4 (2020-03-24)

    • Add ncmultiselector
    • Add ncdirect_cursor_enable() and ncdirect_cursor_disable().