fix CI warnings and errors, fix recalculateFitness()

Author: bhaller

QtSLiM/help/SLiMHelpClasses.html

@@ -1093,9 +1093,10 @@
 <p class="p6">As of SLiM 5.0, this method will read and restore substitutions if they were saved by <span class="s1">outputFull()</span> with its <span class="s1">substitutions=T</span> option.<span class="Apple-converted-space">  </span>This facility is only available when reading binary output from <span class="s1">outputFull()</span>, as chosen by its <span class="s1">binary=T</span> option; otherwise, an error will result.</p>
 <p class="p6">This method can also be used to read tree-sequence information, in the form of single-chromosome <span class="s1">.trees</span> files and multi-chromosome trees archives, as saved by <span class="s1">treeSeqOutput()</span> or generated by the Python <span class="s1">pyslim</span> package.<span class="Apple-converted-space">  </span>Note that the user metadata for a tree-sequence file can be read separately with the <span class="s1">treeSeqMetadata()</span> function.<span class="Apple-converted-space">  </span>Beginning with SLiM 4, the <span class="s1">subpopMap</span> parameter may be supplied to re-order the populations of the input tree sequence when it is loaded in to SLiM.<span class="Apple-converted-space">  </span>This parameter must have a value that is a <span class="s1">Dictionary</span>; the keys of this dictionary should be SLiM population identifiers as <span class="s1">string</span> values (e.g., <span class="s1">"p2"</span>), and the values should be indexes of populations in the input tree sequence; a key/value pair of <span class="s1">"p2", 4</span> would mean that the fifth population in the input (the one at zero-based index <span class="s1">4</span>) should become <span class="s1">p2</span> on loading into SLiM.<span class="Apple-converted-space">  </span>If <span class="s1">subpopMap</span> is non-<span class="s1">NULL</span>, <i>all</i> populations in the tree sequence must be explicitly mapped, even if their index will not change and even if they will not be used by SLiM; the only exception is for unused slots in the population table, which can be explicitly remapped but do not have to be.<span class="Apple-converted-space">  </span>For instance, suppose we have a tree sequence in which population <span class="s1">0</span> is unused, population <span class="s1">1</span> is not a SLiM population (for example, an ancestral population produced by <span class="s1">msprime</span>), and population 2 is a SLiM population, and we want to load this in with population 2 as <span class="s1">p0</span> in SLiM.<span class="Apple-converted-space">  </span>To do this, we could supply a value of <span class="s1">Dictionary("p0", 2, "p1", 1, "p2", 0)</span> for <span class="s1">subpopMap</span>, or we could leave out slot <span class="s1">0</span> since it is unused, with <span class="s1">Dictionary("p0", 2, "p1", 1)</span>.<span class="Apple-converted-space">  </span>Although this facility cannot be used to remove populations in the tree sequence, note that it may <i>add</i> populations that will be visible when <span class="s1">treeSeqOutput()</span> is called (although these will not be SLiM populations); if, in this example, we had used <span class="s1">Dictionary("p0", 0, "p1", 1, "p5", 2)</span> and then we wrote the result out with <span class="s1">treeSeqOutput()</span>, the resulting tree sequence would have six populations, although three of them would be empty and would not be used by SLiM.<span class="Apple-converted-space">  </span>The use of <span class="s1">subpopMap</span> makes it easier to load simulation data that was generated in Python, since that typically uses an id of <span class="s1">0</span>.<span class="Apple-converted-space">  </span>The <span class="s1">subpopMap</span> parameter may not be used with file formats other than tree-sequence files, at the present time; setting up the correct subpopulation ids is typically easier when working with those other formats.<span class="Apple-converted-space">  </span>Note the <span class="s1">tskit</span> command-line interface can be used, like <span class="s1">python3 -m tskit populations file.trees</span>, to find out the number of subpopulations in a tree-sequence file and their IDs.</p>
 <p class="p6">When loading a tree sequence, a crosscheck of the loaded data will be performed to ensure that the tree sequence was well-formed and was loaded correctly.<span class="Apple-converted-space">  </span>When running a Release build of SLiM, however, this crosscheck will only occur the first time that <span class="s1">readFromPopulationFile()</span> is called to load a tree sequence; subsequent calls will not perform this crosscheck, for greater speed when running models that load saved population state many times (such as models that are conditional on fixation).<span class="Apple-converted-space">  </span>If you suspect that a tree sequence file might be corrupted, or might be read incorrectly, running a Debug build of SLiM enables crosschecks after every load.</p>
-<p class="p3">– (void)recalculateFitness([Ni$ tick = NULL])</p>
+<p class="p3">– (void)recalculateFitness([Ni$ tick = NULL]<span class="s5">, [logical$ forceRecalc = T]</span>)</p>
 <p class="p6">Force an immediate recalculation of fitness values for all individuals in all subpopulations.<span class="Apple-converted-space">  </span>Normally fitness values are calculated at a fixed point in each tick, and those values are cached and used until the next recalculation.<span class="Apple-converted-space">  </span>If simulation parameters are changed in script in a way that affects fitness calculations, and if you wish those changes to take effect immediately rather than taking effect at the next automatic recalculation, you may call <span class="s1">recalculateFitness()</span> to force an immediate recalculation and recache.</p>
 <p class="p6">The optional parameter <span class="s1">tick</span> provides the tick for which <span class="s1">mutationEffect()</span> and <span class="s1">fitnessEffect()</span> callbacks should be selected; if it is <span class="s1">NULL</span> (the default), the current tick value for the simulation, <span class="s1">community.tick</span>, is used.<span class="Apple-converted-space">  </span>If you call <span class="s1">recalculateFitness()</span> in an <span class="s1">early()</span> event in a WF model, you may wish this to be <span class="s1">community.tick - 1</span> in order to utilize the <span class="s1">mutationEffect()</span> and <span class="s1">fitnessEffect()</span> callbacks for the previous tick, as if the changes that you have made to fitness-influencing parameters were already in effect at the end of the previous tick when the new generation was first created and evaluated (usually it is simpler to just make such changes in a <span class="s1">late()</span> event instead, however, in which case calling <span class="s1">recalculateFitness()</span> is probably not necessary at all since fitness values will be recalculated immediately afterwards).<span class="Apple-converted-space">  </span>Regardless of the value supplied for <span class="s1">tick</span> here, <span class="s1">community.tick</span> inside callbacks will report the true tick number, so if your callbacks consult that parameter in order to create tick-specific fitness effects you will need to handle the discrepancy somehow.<span class="Apple-converted-space">  </span>(Similar considerations apply for nonWF models that call <span class="s1">recalculateFitness()</span> in a <span class="s1">late()</span> event, which is also not advisable in general.)</p>
+<p class="p6">If <span class="s1">forceRecalc</span> is <span class="s1">T</span> (the default), this method will force the recalculation of all trait values upon which fitness depends.<span class="Apple-converted-space">  </span>(If a trait does not have a direct effect on fitness, it will not be recalculated even when <span class="s1">forceRecalc</span> is <span class="s1">T</span>; use <span class="s1">demandPhenotype()</span> to force recalculation of such traits.)<span class="Apple-converted-space">  </span>If <span class="s1">forceRecalc</span> is <span class="s1">F</span>, trait values will only be recalculated if they are marked as invalid – in other words, if their value is <span class="s1">NAN</span>.<span class="Apple-converted-space">  </span>This option can improve performance by skipping redundant calculations, but can produce out-of-date fitness values if something in the model state has changed such that trait values actually do need to be recalculated even though they are not <span class="s1">NAN</span>.<span class="Apple-converted-space">  </span>This could occur if, for example, if a <span class="s1">mutationEffect()</span> callback’s activation state has been changed.</p>
 <p class="p6">After this call, the fitness values used for all purposes in SLiM will be the newly calculated values.<span class="Apple-converted-space">  </span>Calling this method will trigger the calling of any enabled and applicable <span class="s1">mutationEffect()</span> and <span class="s1">fitnessEffect()</span> callbacks, so this is quite a heavyweight operation; you should think carefully about what side effects might result (which is why fitness recalculation does not just occur automatically after changes that might affect fitness values).</p>
 <p class="p3">– (object&lt;SLiMEidosBlock&gt;$)registerFitnessEffectCallback(Nis$ id, string$ source, [Nio&lt;Subpopulation&gt;$ subpop<span class="s9"> </span>= NULL], [Ni$ start = NULL], [Ni$ end = NULL])</p>
 <p class="p6">Register a block of Eidos source code, represented as the <span class="s1">string</span> singleton <span class="s1">source</span>, as an Eidos <span class="s1">fitnessEffect()</span> callback in the current simulation (specific to the target species), with an optional subpopulation <span class="s1">subpop</span> (which may be an <span class="s1">integer</span> identifier, or <span class="s1">NULL</span>, the default, to indicate all subpopulations), and optional <span class="s1">start</span> and <span class="s1">end</span> ticks all limiting its applicability.<span class="Apple-converted-space">  </span>The script block will be given identifier <span class="s1">id</span> (specified as an <span class="s1">integer</span>, or as a <span class="s1">string</span> symbolic name such as <span class="s1">"s5"</span>); this may be <span class="s1">NULL</span> if there is no need to be able to refer to the block later.<span class="Apple-converted-space">  </span>The registered callback is added to the end of the list of registered <span class="s1">SLiMEidosBlock</span> objects, and is active immediately; it <i>may</i> be eligible to execute in the current tick.<span class="Apple-converted-space">  </span>The new <span class="s1">SLiMEidosBlock</span> will be defined as a global variable immediately by this method, and will also be returned by this method.</p>

SLiMgui/SLiMHelpClasses.rtf

@@ -10487,7 +10487,7 @@ When loading a tree sequence, a crosscheck of the loaded data will be performed
 \f4\fs20  is called to load a tree sequence; subsequent calls will not perform this crosscheck, for greater speed when running models that load saved population state many times (such as models that are conditional on fixation).  If you suspect that a tree sequence file might be corrupted, or might be read incorrectly, running a Debug build of SLiM enables crosschecks after every load.\
 \pard\pardeftab397\li720\fi-446\ri720\sb180\sa60\partightenfactor0
 
-\f3\fs18 \cf0 \'96\'a0(void)recalculateFitness([Ni$\'a0tick\'a0=\'a0NULL])\
+\f3\fs18 \cf0 \'96\'a0(void)recalculateFitness([Ni$\'a0tick\'a0=\'a0NULL]\cf2 , [logical$\'a0forceRecalc\'a0=\'a0T]\cf0 )\
 \pard\pardeftab720\li547\ri720\sb60\sa60\partightenfactor0
 
 \f4\fs20 \cf2 Force an immediate recalculation of fitness values for all individuals in all subpopulations.  Normally fitness values are calculated at a fixed point in each tick, and those values are cached and used until the next recalculation.  If simulation parameters are changed in script in a way that affects fitness calculations, and if you wish those changes to take effect immediately rather than taking effect at the next automatic recalculation, you may call 
@@ -10526,7 +10526,30 @@ The optional parameter
 \f4\fs20  in a 
 \f3\fs18 late()
 \f4\fs20  event, which is also not advisable in general.)\
-After this call, the fitness values used for all purposes in SLiM will be the newly calculated values.  Calling this method will trigger the calling of any enabled and applicable 
+\pard\pardeftab720\li547\ri720\sb60\sa60\partightenfactor0
+\cf2 If 
+\f3\fs18 forceRecalc
+\f4\fs20  is 
+\f3\fs18 T
+\f4\fs20  (the default), this method will force the recalculation of all trait values upon which fitness depends.  (If a trait does not have a direct effect on fitness, it will not be recalculated even when 
+\f3\fs18 forceRecalc
+\f4\fs20  is 
+\f3\fs18 T
+\f4\fs20 ; use 
+\f3\fs18 demandPhenotype()
+\f4\fs20  to force recalculation of such traits.)  If 
+\f3\fs18 forceRecalc
+\f4\fs20  is 
+\f3\fs18 F
+\f4\fs20 , trait values will only be recalculated if they are marked as invalid \'96 in other words, if their value is 
+\f3\fs18 NAN
+\f4\fs20 .  This option can improve performance by skipping redundant calculations, but can produce out-of-date fitness values if something in the model state has changed such that trait values actually do need to be recalculated even though they are not 
+\f3\fs18 NAN
+\f4\fs20 .  This could occur if, for example, if a 
+\f3\fs18 mutationEffect()
+\f4\fs20  callback\'92s activation state has been changed.\
+\pard\pardeftab720\li547\ri720\sb60\sa60\partightenfactor0
+\cf2 After this call, the fitness values used for all purposes in SLiM will be the newly calculated values.  Calling this method will trigger the calling of any enabled and applicable 
 \f3\fs18 mutationEffect()
 \f4\fs20  and 
 \f3\fs18 fitnessEffect()

VERSIONS

@@ -123,6 +123,7 @@ multitrait branch:
 		note that there is no concept of independent dominance for the hemizygous dominance coefficient, since hemizygous mutations are present in only one copy by definition
 	switch fitness calculation over to being based upon the calculated values of traits, rather than directly upon mutations
 	eliminate overhead for setting up fitness buffers in neutral WF models; this should provide a significant speedup for such models, if they don't use a mateChoice() callback
+	extend recalculateFitness() with [logical$ forceRecalc = T] option for backward compatibility, but allowing trait value recalculation not to be forced
 
 
 version 5.1 (Eidos version 4.1):

core/community.cpp

@@ -2894,7 +2894,7 @@ bool Community::_RunOneTickWF(void)
 				if (is_explicit_species_)
 					gSLiMScheduling << "\tfitness recalculation: species " << species->name_ << std::endl;
 #endif
-				species->RecalculateFitness();
+				species->RecalculateFitness(/* p_force_trait_recalculation */ false);
 
 				executing_species_ = nullptr;
 			}
@@ -3197,7 +3197,7 @@ bool Community::_RunOneTickNonWF(void)
 				if (is_explicit_species_)
 					gSLiMScheduling << "\tfitness recalculation: species " << species->name_ << std::endl;
 #endif
-				species->RecalculateFitness();
+				species->RecalculateFitness(/* p_force_trait_recalculation */ false);
 
 				executing_species_ = nullptr;
 			}

core/haplosome.cpp

@@ -2858,7 +2858,7 @@ EidosValue_SP Haplosome_Class::ExecuteMethod_addNewMutation(EidosGlobalStringID
 	// for the singleton case for each of the parameters, get all the info
 	MutationType *singleton_mutation_type_ptr = SLiM_ExtractMutationTypeFromEidosValue_io(arg_muttype, 0, &community, species, method_name.c_str());		// SPECIES CONSISTENCY CHECK
 
-	slim_effect_t singleton_selection_coeff = (arg_effect ? (slim_effect_t)arg_effect->NumericAtIndex_NOCAST(0, nullptr) : 0.0);
+	slim_effect_t singleton_selection_coeff = (arg_effect ? (slim_effect_t)arg_effect->NumericAtIndex_NOCAST(0, nullptr) : (slim_effect_t)0.0);
 
 	slim_position_t singleton_position = SLiMCastToPositionTypeOrRaise(arg_position->IntAtIndex_NOCAST(0, nullptr));
 

core/mutation.cpp

@@ -78,8 +78,8 @@ mutation_type_ptr_(p_mutation_type_ptr), position_(p_position), subpop_index_(p_
 
 		// FIXME MULTITRAIT: This constructor needs to change to have a whole vector of trait information passed in, for effect and dominance
 		// for now we use the values passed in for trait 0, and make other traits neutral
-		slim_effect_t effect = (trait_index == 0) ? p_selection_coeff : 0.0;
-		slim_effect_t dominance = (trait_index == 0) ? p_dominance_coeff : 0.5;		// can be NAN
+		slim_effect_t effect = (trait_index == 0) ? p_selection_coeff : (slim_effect_t)0.0;
+		slim_effect_t dominance = (trait_index == 0) ? p_dominance_coeff : (slim_effect_t)0.5;		// can be NAN
 		slim_effect_t hemizygous_dominance = mutation_type_ptr_->DefaultHemizygousDominanceForTrait(trait_index);	// FIXME MULTITRAIT: This needs to come in from outside, probably
 
 		traitInfoRec->effect_size_ = effect;
@@ -305,8 +305,8 @@ mutation_type_ptr_(p_mutation_type_ptr), position_(p_position), subpop_index_(p_
 
 		// FIXME MULTITRAIT: This constructor needs to change to have a whole vector of trait information passed in, for effect and dominance
 		// for now we use the values passed in for trait 0, and make other traits neutral
-		slim_effect_t effect = (trait_index == 0) ? p_selection_coeff : 0.0;
-		slim_effect_t dominance = (trait_index == 0) ? p_dominance_coeff : 0.5;		// can be NAN
+		slim_effect_t effect = (trait_index == 0) ? p_selection_coeff : (slim_effect_t)0.0;
+		slim_effect_t dominance = (trait_index == 0) ? p_dominance_coeff : (slim_effect_t)0.5;		// can be NAN
 		slim_effect_t hemizygous_dominance = mutation_type_ptr_->DefaultHemizygousDominanceForTrait(trait_index);	// FIXME MULTITRAIT: This needs to come in from outside, probably
 
 		traitInfoRec->effect_size_ = effect;

core/population.cpp

@@ -5241,7 +5241,7 @@ void Population::AddTallyForMutationTypeAndBinNumber(int p_mutation_type_index,
 }
 #endif
 
-void Population::RecalculateFitness(slim_tick_t p_tick)
+void Population::RecalculateFitness(slim_tick_t p_tick, bool p_force_trait_recalculation)
 {
 	// calculate the fitnesses of the parents and make lookup tables; the main thing we do here is manage the mutationEffect() callbacks
 	// as per the SLiM design spec, we get the list of callbacks once, and use that list throughout this stage, but we construct
@@ -5438,7 +5438,7 @@ void Population::RecalculateFitness(slim_tick_t p_tick)
 		std::vector<SLiMEidosBlock*> no_callbacks;
 
 		for (std::pair<const slim_objectid_t,Subpopulation*> &subpop_pair : subpops_)
-			subpop_pair.second->UpdateFitness(no_callbacks, no_callbacks, p_direct_effect_trait_indices);
+			subpop_pair.second->UpdateFitness(no_callbacks, no_callbacks, p_direct_effect_trait_indices, p_force_trait_recalculation);
 	}
 	else
 	{
@@ -5466,7 +5466,7 @@ void Population::RecalculateFitness(slim_tick_t p_tick)
 			}
 
 			// Update fitness values, using the callbacks
-			subpop->UpdateFitness(subpop_mutationEffect_callbacks, subpop_fitnessEffect_callbacks, p_direct_effect_trait_indices);
+			subpop->UpdateFitness(subpop_mutationEffect_callbacks, subpop_fitnessEffect_callbacks, p_direct_effect_trait_indices, p_force_trait_recalculation);
 		}
 	}
 

core/population.h

@@ -221,7 +221,7 @@ class Population
 	Individual *(Subpopulation::*GenerateIndividualCloned_TEMPLATED)(Individual *p_parent) = nullptr;
 
 	// Recalculate all fitness values for the parental generation, including the use of mutationEffect() callbacks
-	void RecalculateFitness(slim_tick_t p_tick);
+	void RecalculateFitness(slim_tick_t p_tick, bool p_force_trait_recalculation);
 
 	// Scan through all mutation runs in the simulation and unique them
 	void UniqueMutationRuns(void);