Changes from 3.7.3 to 3.7.4

Additions, changes

• Bump year to 2025

• AddBorders and LetterBox: add transient filtering See AddBorders.

• new filter: MultiOverlay. Bulk copy-paste from clips. See MultiOverlay.

• Animate: Custom function option. See Animate.

• Animate: more precise granularity for integer interpolation, allow 64 bit input

• Animate: add proper rounding for integer interpolation

• Resizers: add "force" integer parameter to force the resizing process even if Avisynth decides it is not needed

• Resizers: respect _ChromaLocation, and "placement" parameter which to adjust chroma locations during resizing

• Resizers: add "keep_center" parameter to disable pixel center position adjustments

• Resizers: allow arbitrary dimensions; the filter "support" size does not limit usability, no more "image height is too small for this resizing method"-like error messages.

• GaussResize: add "b" and "s" parameters See Resize Filters.

• DirectShowSource new parameter utf8 for utf8 filename support

• "propShow" font, text_color, halo_color, bold new parameters for custom style.

• "propShow" (#366): x, y, align new parameters for custom positioning.

• "propShow" props new parameter for filtering the name of property list with wildcard and regex support.

  See propShow.

• "propDelete" and "propCopy": props accepts wildcard and regex for filtering the name of property list.

  See Internal functions: frame properties.

• "Info": cpu new parameter to disable showing CPU capabilities

• "Info" (#366): x, y, align new parameters for custom positioning

• Fix #368 Make proper vertical alignment for multiline text in Subtitle and Text when vertical alignment is set to bottom or center.

• Studio RGB (narrow, limited) range will now be recognized (through _ColorRange=1) and utilized in conversions to and from RGB, such as in GreyScale, ConvertToY, ConvertToYUVxxx, or ConvertToRGB32 with PC.xxx matrix which now keeps the limited/narrow-rangeness of the source. See also BugFix section.

• #392 "break" and "continue" in for-next and while loops

• Add "ArraySort" for sorting simple bool, numeric or string arrays

• SoundTouch: Increased the max number of channels to 32ch from the old limit of 16ch.

• Posix: Detect additional plugindirs from LD_LIBRARY_PATH environment variable

• #413 Add ListAutoloadDirs() script function returing a LF delimited string with directory list.

• SubTitle to accept real LF (\r) or CR LF (\r\n) control characters for line break.

• #422 Give meaningful and stopping error message when a bad C++ plugin is throwing exception during autoload (e.g. wrong function signatures, crash because of using unsupported interface calls). This kind of loading error is treated similarly fatal like mixing 32 and 64 bit plugins.

• #413 Add 64 bit data types (https://github.com/AviSynth/AviSynthPlus/issues/423)

  • (technical) Value (AVSValue) types i and f are kept and still mean 32 bit integer and float values.

  • (technical) New value types l and d added to mark 64 bit integer and double types.

  • Integer decimal constants are automatically promoted to long (int64_t) if they do not fit into integer. Integer decimal constants are automatically promoted to 64 bit double if they do not even fit into a 64 bit long. (But values stored as 'only' integers behave as 64 bit long in math operations).

  • New syntax element L (or l) suffix for hexadecimal constants for interpreting 64 bit data ($800000001010L) Hexadecimal literals without L suffix still remain 32 bit integers. This ensures compatibility so that color constants exceeding $7FFFFFFF (such as $FFFFFFFF) are still interpreted as 32-bit integers and not as 64-bit long integers.

  • Floating point constants (literals) stored as 64 bit double (formerly only with 32 bit float precision)

  • IsInt() returns true for any 32 or 64 bit integer

  • IsLongStrict() returns true only if underlying variable is long (64 bit integer)

  • IsFloat() returns true for any 32 or 64 bit floating point content (float or double and for any integers - as before)

  • IsFloatfStrict() returns true only if value is exactly a 32 bit single-precision float.

  • Float() by default casts to double, but 32-bit floating point numbers are kept in 32-bit float format (no automatic float-to-double conversion).

  • New Double() converts always to 64 bit double

  • New Floatf() converts always to 32 bit float

  • New Long() converts always to 64 bit long

  • New IntI() converts always to 32 bit integer

  • Function Pi returns real double precision constant.

  • VersionNumber() still returns a 32-bit float, but the exact value is adjusted slightly for compatibility reasons. This is because the script constant 2.6 is now stored in 64-bit double precision, which can cause issues when compared to a 32-bit float version of 2.6. For example, to ensure 2.6 (double) >= 2.6f (float) and avoid breaking scripts that contain IsAvs26 = VersionNumber() >= 2.6 .

  • Floating point mathematical functions (Sin, etc..) return double precision results.

  • Floating point arithmetic (addition, subtraction, multiplication, division) uses double precision, except when both operands are 32-bit floats, in which case the result is also a 32-bit float. (32-bit float / 32-bit float results in a 32-bit float)

  • Integer mathematical operation operate on 64 bit data. When result is within a 32 bit integer range, it is stored as 32 bit integer internally, which is automatically promoted to 64 bit long if needed.

  • For-loops are using 64 bit integer for initial value, step and end-value.

  • Animate: using 64 bit precision inside.

  • Bit-related functions now have distinctly named 64-bit integer versions, while the old ones continue to work with plain 32-bit integers.

    • BitAnd64, BitNot64, BitOr64, BitXor64,

    • bitshl64, bitsal64: shift left (providing two versions; though arithmetic and logical is the same)

    • bitshr64, bitsar64: shift right logical and arithmetic

    • bitrol64, bitror64: 64 bit rotation left and right

    • BitChg64, BitClr64, BitSet64, BitTst64, change, clear, set, test

    • bitsetcount64 counts the set bits (or sum of set bits) of one or more parameter values.

  • New: HexValue64() for 64-bit long result. The original HexValue() keeps returning 32 bit integer so HexValue("FFFFFFFF") is still -1, but HexValue64("FFFFFFFF") will be 4294967295 (64 bit number)

  • Integer and float frame property read and writes work on real 64 bit integers and doubles.

  • Formatting functions (String(), etc.) to 64 bit data type aware.

  • AudioLength() now returns 64 bit integer

  • AudioLengthF() returns data to double instead of 32 bit float (though it's now useless and still can lose precision).

  • Function parameter types (function signature) keep i and f, but accept 64 bit long and double precision parameters transparently.

  • String length can exceed the value of a 32 bit int. StrLen() returns long if int cannot hold the length. LeftStr, RightStr, MidStr length parameters accept 64 bit long values.

  • Compatibility: 64 bit parameter values are automatically replaced with 32 bit int and 32 bit float if Avisynth 2.5 or old (pre-V11, non-64-bit aware) C interface is detected. Such plugins will see the good old 32 bit data.

  • Backport the frame property changes from VapourSynth API4

    • new saturated frame property getters on API returning only 32 bit data: propGetIntSaturated and propGetFloatSaturated

    • A frame property with data type can hold a string or real binary data. To distinguish between the two kinds, a hint can be set with a new API version of propSetDataH, which helps visualizing and later: exporting.

    • propShow displays binary buffer instead of string if the internal subtype is set to Binary. The buffer length and the first (up to) 16 bytes are displayed like this: Binary data. Length= xx [00,01,DD, ...] A string content is displayed otherwise.

    • Since propSet script function in Avisynth is string-only, it sets the hint to DATATYPEHINT_UTF8 (utf8 just hinting that this is a string)

• WinInstaller: add InnoSetup script for Windows on ARM

• buildsystem: restrict use of MSVC to x86(-64) only

• avs/config.h: add defines to separate X86 and ARM Windows variants

• PluginManager: only enable +GCC plugindir registry entries on X86

• PluginManager: indent cosmetics for clarity

• Restore AVS_VERSION define

Build environment, Interface

• add LOONGARCH support

• Use system installs of DevIL and SoundTouch on all platforms, remove in-tree binaries/code

• avisynth.h: add ListAutoloadDirs() to internal interface declarations

• CMakeList.txt to accept Intel C++ Compiler 2025

• V11 interface: new 64 bit related AVSValue get and set function in C++ and C interface.

• V11 interface: C Interface: implement API for all getter/setter/typecheck for AVS_Value

• V11 interface: C interface supports Avisynth+ deep-copy dynamic arrays.

• Added optional C plugin init function: to enable full 64 bit data to C plugins, they should implement avisynth_c_plugin_init2.

• V11: C interface add avs_add_function_r as an alternative to avs_add_function, allowing the callback to return the result via a by-reference AVS_Value parameter instead of returning the AVS_Value as a struct. (Use case from Python) See avs_add_function_r

• V11 interface: add saturated int (int64->int) and float (double->float) frame property reading function to IScriptInterface: propGetIntSaturated, propGetFloatSaturated and prop_get_int_saturated, prop_get_float_saturated to C interface, like in VapourSynth API 4.

• V11 interface: "hint" to buffer/string frame properties, like in Vapoursynth API 4. IScriptInterface: propGetDataTypeHint, propSetDataH. C interface: avs_prop_get_data_type_hint, avs_prop_set_data_h. Equivalent functions: VSAPI4.mapGetDataTypeHint = Avisynth.propGetDataTypeHint, VSAPI4.mapSetData = Avisynth.propSetDataH, VSAPI4.mapSetData3 = Avisynth.propSetData.

• V11: New enum in headers: AVSPropDataTypeHint (VSAPI4: VSDataTypeHint)

• Background modification: env->SaveString can store longer strings than INT_MAX if len is -1 (autodetect length by null termination). Even on 32 bit systems size_t can exceed INT_MAX. (nevertheless, the length parameter - when is given - is still int type)

• Shibatch: check for GCC or Clang for MulDiv under MinGW

• plugins/CMakeLists.txt: restrict VDubFilter to x86(-64)

• plugin/CMakeLists: add CMake detection fallback for SoundTouch

• CMakeLists.txt: use extended policy version requirements

Bugfixes

• Fix ConvertBits C 16->8 bit (x+round, then bitshift) which turned 0xFFFF into 256 which is 0 (wrong)

• Fix ConvertToRGB48/64 debug assert which passed less than adequate parameters to an internal PlanarRGb converter

• Fix: Resizers chroma shift if not chroma is not center-positioned (respect _ChromaLocation, and "placement" parameter)

• Fix #429 CPU-CUDA transfer would copy less bytes than vfb data_size

• Fix: ScriptClip and other runtime functions / frame prop read would crash if called from Avs2.5 or Pre-v11 C IScriptEnvironment, which would happen is there is an old C plugin within ScriptClip The C part was an intermediate development issue, but for Avs 2.5 plugins probably it always crashed.

• Fix: audio cache would overflow over int sized count and position differences

• Fix: "Info" overflowed and displayed negative hours/minutes/audio length at longer clips, use int64

• Fix "SetLogParams" defaults - mentioned in #391

• Fix corrupt Turn functions when a planar RGB turn would be followed by a YUV Turn. Regression since TurnXXXX supports planar RGB (2016.08.23; probably since r2081 commit dba954e2de0c9c6218d17fc5c4974f4c28b627c3)

• Fix #386: Interleave to call plugin destructor like StackXXXX (memory leak in case of script errors) additional code reorg in r4192

• Fix #384: swapped ShowGreen/ShowBlue for planar RGB sources

• Fix: allow use of "local" in ConditionalSelect string version (fixed wrong function signature)

• "Info" now can display a line which is only partially visible (instead of not showing it at all)

• "Text" use "lsp" parameter the same way as in SubTitle: in 1/8 pixel units, not in 1 pixels.

• "Text" vertical alignment position would be wrong for multiline strings containing even number of lines.

• Fix #365: Avisynth 2.5 plugins when NICE_FILTER would crash with "invalid response to CACHE_GETCHILD_AUDIO_MODE". Regression in 3.7.3 reintroduced audio cache.

• Fix #370: array size assert error in ConvertToYUY2 when internally ConvertToYUV422 is called

• Leave _ColorRange frame property as-is, when using matrix names "PC.709" or "PC.601", for example in ConvertToRGB32. This, along with narrow-range RGB support, also fixes an old glitch, when a limited range YUV Source (like ColorBarsHD) would be converted to a full-range RGB (RGBs were assumed full range).

• Narrow-range RGB support fixes the minor +/-1 bit error in YUV to RGB (and back) matrix conversions for such RGB sources and targets. This error would accumulate if conversions were applied consecutively or mixed with other external conversions.

• Fix FadeIn, which behaved like FadeIn2

• Fix #418: ShowSMTPE/ShowTime alignment - NO_WIN_GDI case (e.g. Posix)

• Fix #421: add safety checks to SuperEQ to match SSRC

• Fix #426: An autoloaded function with multiple parameter versions would produce a named parameter error if unnamed parameters were given before the named parameter. For example: TimeStretch(a_clip, rate=100.1)

Optimizations

• avoid storing duplicated strings in internal string heap (related to issue #389)

• Expr: rewritten the C (non-Intel-JIT) path to support vectorization, if the compiler is capable. Useful for non-Intel platforms where the (Intel SSE2-AVX2) JIT compiler does not work. Expect 3-20x speedup compared to the old method.

• Expr: implement tan in JITasm. Expect ~6-15x speed up for an expression like "sxr 2 * 1 - 3.14159254 * 1 * tan 10 * 128 +"

• Resizers C implementation: more vectorizer compiler friendly code (1.5 - 2.5 speed, still slooow)

• Quicker SSE2 horizontal and vertical resizer

• [Un-optimization]: minor speed decrease in other resizers' performance, due to healing a hidden possibility which would allow over-addressing the scan-lines and frame buffer. No wonder the old code, which checked nothing, did well. IMHO the code is still quick.

Documentation

• Correct building DirectShowSource prerequisites (Release_MBCS)

• Update "DirectShowSource" with utf8 parameter

• Update "Info"

• Update rst docs with control structs if/else/for/while

• Update "ShowTime", "ShowSMPTE", "ShowFrameNumber" section with 3.7.3 changes

• Update most items at Syntax and internal functions sections, add arrays, function objects, escaped string literals, multithreading, frame properties, debug functions

• Add if-else, do-while, for-next, break and continue

• Update Import (add utf8)

• update Conditional filters, Runtime functions

• update ShowAlpha/Red/...

• Update Intel 2024 compiler build process

• Add autoload helper functions to rst docs, update content and bring syntax_plugins.rst to new format

• update compile_avsplus.rst online docs with Intel C++ Compiler 2025

• Interface V11 changes: see What's new in the API V11 for more details.

• Document 64 bit capable functions and changes in old ones in millions of other rst pages.

• Document C interface (in progress)

• Revamped resizer section Resize filters

• Add instructions for building or using external system installs of SoundTouch and DevIL

• and so on...

Please report bugs at github AviSynthPlus page - or - Doom9's AviSynth+ forum

$Date: 2025/03/14 14:09:00 $