--- linuxsampler/trunk/ChangeLog 2020/01/23 18:15:37 3724 +++ linuxsampler/trunk/ChangeLog 2021/06/19 15:27:02 3964 @@ -1,5 +1,14 @@ Version SVN trunk (?) + * GigaStudio/Gigasampler format engine: + - Use libgig's new IO-per-thread feature to avoid file I/O concurrency + issues with .gig file. + - Use libgig's new method gig::File::GetSample() with method + gig::File::CountSamples() instead of now deprecated + gig::File::GetFirstSample() / gig::File::GetNextSample() pair. + +Version 2.2.0 (9 May 2021) + * general changes: - Fixed compiler error in Pool.h. - Require C++14 compiler support. @@ -49,6 +58,31 @@ entry/increment/decrement CC messages. - Fixed crash when editing key groups (a.k.a. exclusive groups) with an instrument editor. + - Fixed playing noise at the end of a disk stream under certain conditions. + - Fixed potential crash with high pitch at end of RAM playback. + - Added configure checks whether C++ compiler supports so called + 'designated initializers'. + - Added global backtraceAsString() function for debugging purposes + (POSIX systems only, no Windows implementation yet). + - Thread class: Added new methods pushCancelable(), popCancelable(), name(), + nameOfCaller() and setNameOfCaller(). + - configure: Always check for availability of pthread_testcancel() and use + it if so (previously it was only used if explicitly enabled by configure + option). + - Thread class: raise compiler warning if pthread_testcancel() is not + available, because it can lead to deadlocks. + - Mutex class: Implemented optional bug detection and deadlock debugging + features. + - Ref<> class is now thread safe (as lock-free and wait-free + implementation). + - Fixed compilation error with Bison >= 3.6.0. + - Fixed rare crash when switching between instruments that had instrument + scripts loaded. + - Allow EffectChain::InsertEffect() as alternative to + EffectChain::AppendEffect() (i.e. with index equal to current chain size). + - ConstCapacityArray class: Fixed bug in remove() method. + - Fixed crash on certain MIDI RPN and NRPN messages that occurred if no + real-time instrument script was loaded. * Real-time instrument scripts: - Added method ScriptVM::setExitResultEnabled() which allows to @@ -242,6 +276,51 @@ "change_amp_lfo_depth()", "change_cutoff_lfo_depth()" and "change_pitch_lfo_depth()", to allow e.g. passing value 2000000 to double the LFO frequency / resonance / cutoff frequency. + - Reduced code involved for parsing NKSP variable declarations by using 2 + generalized grammar rules with optional components (i.e. optional value + assignment and optional variable qualifiers) instead of previous linear + combinations of grammar rules, which also fixes some edge case. + - NKSP language parser: track code block locations also by a) raw byte + position in NKSP source code string and b) length of that code block in + bytes (along to the already existing code block tracking by line/column); + and added appropriate new members to public C++ API classes like + 'firstByte' and 'lengthBytes' to class VMSourceToken and struct CodeBlock. + - NKSP language: Added support for 'patch' variable qualifier (capture their + locations and allow replacement of their assignment expression via VM C++ + API). + - Fixed incorrect EG times if standard measuring unit was passed to built-in + functions "change_attack()", "change_decay()" or "change_release()". + - Fixed intermediate function result values never having reflected any + standard measuring unit type. + - Fixed re-entrant issue with function calls which caused wrong result + values if the same function was called multiple times in a term + (specifically if metric prefixes were used). + - NKSP language: Just throw a warning, not an error if an array variable of + size zero was declared. + - NKSP language: Allow omitting explicit array size on array variable + declaration if combined with immediate value assignment + (e.g. declare %foo[] := ( 1, 2, 3 ) ). + - NKSP language: emit warning if an array variable was declared with bigger + array size than amount of initial values been assigned, and initialize the + missing array elements with zero in this case. + - Fixed built-in NKSP exit() function to behave as return statement. + - NKSP parser: Fixed crash if unary '-' operator was used on a non-number + data type. + - NKSP VM: Fixed the "release" handler being executed too often under + certain situations if polyphonic data was passed from "note" handler to + "release" handler. + - Fixed memory leak in NKSP parser: string tokens were allocated as C + strings and never freed. + - Fixed incorrect polyphonic data transfer from wrong note handler to + release handler; which also fixes no release handler being executed + sometimes, and due to the latter it also fixes potential crashes as some + polyphonic script events were never released and the engine hence was + running out of free script events. + - Fixed crash when handling array results of built-in functions + (e.g. "num_elements(by_marks($MARK_1)"). + - NKSP VM: Fixed dead note events not been removed from user's event groups + (i.e. causing incorrect behaviour of built-in NKSP functions by_marks() + and set_event_mark() after max. size of event group list was exhausted). * test cases: - Fixed compiler errors in test cases. @@ -267,6 +346,7 @@ "sin()", "cos()", "tan()", "asin()", "acos()", "atan()". - Added NKSP test cases for unary '+' operator. - Added NKSP test cases for core built-in variables. + - Added thorough NKSP test cases for variable declarations. * GigaStudio/Gigasampler format engine: - LFOTriangleIntMath and LFOTriangleIntAbsMath: Fixed FlipPhase=true @@ -289,10 +369,15 @@ - If minimum cutoff is defined, remap cutoff controller CC values to really span the entire possible cutoff range (and avoiding a dead controller zone). + - Implemented support for loading instrument scripts with 'patch' variables + (by patching these script variables with potentially overridden ones by + gig Instrument). * SFZ format engine: - Fixed support for regions with loccN/hiccN conditions on more than one MIDI controller. + - Fixed internal opcode 'xfin_hivel' falsely being set by user opcode + 'xfin_lovel' (patch by Christian Czezatke). * Benchmarks: - Fixed benchmarks/triang.cpp falsely having favoured "int math abs"