Project

Profile

Help

Bug #2321

closed

Function inlining

Added by Michael Kay about 9 years ago. Updated almost 5 years ago.

Status:
Closed
Priority:
Normal
Assignee:
Category:
Internals
Sprint/Milestone:
-
Start date:
2015-02-28
Due date:
% Done:

100%

Estimated time:
Legacy ID:
Applies to branch:
9.9, trunk
Fix Committed on Branch:
9.9, trunk
Fixed in Maintenance Release:
Platforms:

Description

We should do some experiments with function inlining. (a) the limit on expression size for inlining is fixed at 15, which seems rather low. BaseX defaults the limit to 100, and makes it configurable. (b) we don't inline a function if it contains any calls to user-defined functions. I think this is to prevent infinite recursion. But it doesn't seem necessary: the expansion would stop when we hit the size limit.


Files

taskBug.xsl (1.71 KB) taskBug.xsl Debbie Lockett, 2019-01-16 13:09
taskBug2.xsl (3.24 KB) taskBug2.xsl Debbie Lockett, 2019-01-16 19:28
Actions #1

Updated by Michael Kay over 5 years ago

  • Project changed from 4 to Saxon
  • Description updated (diff)
  • Status changed from New to Resolved
  • Applies to branch 9.9 added
  • Fix Committed on Branch 9.9 added

I have added a configuration option for this (on the 9.9 branch) and I'm setting the default value to 100 rather than 15, to see how we get on.

Handling recursive functions is trickier: if we simply remove the check on calling other user-defined functions, then the compiler goes into infinite recursion.

I haven't found a test case where it makes a measurable difference. The real benefit of inlining comes when the inlined function is capable of further optimization (perhaps even eager compile-time evaluation) as a result of more being known about the supplied arguments.

Actions #2

Updated by O'Neil Delpratt over 5 years ago

  • Status changed from Resolved to Closed
  • % Done changed from 0 to 100
  • Fixed in Maintenance Release 9.9.0.2 added

Bug fix applied to the Saxon 9.9.0.2 maintenance release.

Actions #3

Updated by Debbie Lockett over 5 years ago

Reopening this bug. Have a stylesheet which ran OK with 9.9.0.1, but not with 9.9.0.2+, and it seems that the problem is this increase for the threshold for function inlining.

See attached stylesheet taskBug.xsl. Transform with initial template "ex1" (i.e. -it:ex1 -xsl:taskBug.xsl)

Stack trace:

Exception in thread "main" java.lang.StackOverflowError
	at net.sf.saxon.expr.VariableReference.recomputeInLoop(VariableReference.java:236)
	at net.sf.saxon.expr.VariableReference.typeCheck(VariableReference.java:228)
	at net.sf.saxon.expr.Operand.typeCheck(Operand.java:191)
	at net.sf.saxon.expr.Expression.typeCheckChildren(Expression.java:547)
	at net.sf.saxon.expr.SystemFunctionCall.typeCheck(SystemFunctionCall.java:87)
	at net.sf.saxon.expr.Operand.typeCheck(Operand.java:191)
	at net.sf.saxon.expr.CardinalityChecker.typeCheck(CardinalityChecker.java:99)
	at net.sf.saxon.expr.Operand.typeCheck(Operand.java:191)
	at net.sf.saxon.expr.Expression.typeCheckChildren(Expression.java:547)
	at net.sf.saxon.expr.SystemFunctionCall.typeCheck(SystemFunctionCall.java:87)
	at net.sf.saxon.expr.Operand.typeCheck(Operand.java:191)
	at net.sf.saxon.expr.CardinalityChecker.typeCheck(CardinalityChecker.java:99)
	at net.sf.saxon.expr.Operand.typeCheck(Operand.java:191)
	at net.sf.saxon.expr.ItemChecker.typeCheck(ItemChecker.java:100)
	at net.sf.saxon.expr.Operand.typeCheck(Operand.java:191)
	at net.sf.saxon.expr.Expression.typeCheckChildren(Expression.java:547)
	at net.sf.saxon.expr.SystemFunctionCall.typeCheck(SystemFunctionCall.java:87)
	at net.sf.saxon.expr.Operand.typeCheck(Operand.java:191)
	at net.sf.saxon.expr.LetExpression.typeCheck(LetExpression.java:154)
	at net.sf.saxon.expr.instruct.UserFunction.typeCheck(UserFunction.java:530)
	at com.saxonica.functions.hof.UserFunctionReference.typeCheck(UserFunctionReference.java:78)
	at net.sf.saxon.expr.Operand.typeCheck(Operand.java:191)
	at net.sf.saxon.expr.Expression.typeCheckChildren(Expression.java:547)
	at com.saxonica.functions.hof.PartialApply.typeCheck(PartialApply.java:72)
	at net.sf.saxon.expr.Operand.typeCheck(Operand.java:191)
	at com.saxonica.functions.hof.FunctionSequenceCoercer.typeCheck(FunctionSequenceCoercer.java:73)
	at net.sf.saxon.expr.Operand.typeCheck(Operand.java:191)
	at net.sf.saxon.expr.Expression.typeCheckChildren(Expression.java:547)
	at net.sf.saxon.expr.FunctionCall.typeCheck(FunctionCall.java:169)
	at net.sf.saxon.expr.UserFunctionCall.typeCheck(UserFunctionCall.java:387)
	at net.sf.saxon.expr.instruct.UserFunction.typeCheck(UserFunction.java:530)
	at com.saxonica.functions.hof.UserFunctionReference.typeCheck(UserFunctionReference.java:78)
	at net.sf.saxon.expr.Operand.typeCheck(Operand.java:191)
	at com.saxonica.functions.hof.FunctionSequenceCoercer.typeCheck(FunctionSequenceCoercer.java:73)
	at net.sf.saxon.expr.Operand.typeCheck(Operand.java:191)
	at net.sf.saxon.expr.Expression.typeCheckChildren(Expression.java:547)
	at com.saxonica.functions.hof.PartialApply.typeCheck(PartialApply.java:72)
	at net.sf.saxon.expr.Operand.typeCheck(Operand.java:191)
	at net.sf.saxon.expr.Expression.typeCheckChildren(Expression.java:547)
	at net.sf.saxon.expr.SystemFunctionCall.typeCheck(SystemFunctionCall.java:87)
	at net.sf.saxon.expr.Operand.typeCheck(Operand.java:191)
	at net.sf.saxon.expr.Expression.typeCheckChildren(Expression.java:547)
	at net.sf.saxon.expr.instruct.Block.typeCheck(Block.java:648)
	at net.sf.saxon.expr.Operand.typeCheck(Operand.java:191)
	at net.sf.saxon.expr.Expression.typeCheckChildren(Expression.java:547)
	at net.sf.saxon.expr.SystemFunctionCall.typeCheck(SystemFunctionCall.java:87)
	at net.sf.saxon.expr.Operand.typeCheck(Operand.java:191)
	at net.sf.saxon.expr.LetExpression.typeCheck(LetExpression.java:154)
	at com.saxonica.ee.optim.OptimizerEE.tryInlineFunctionCall(OptimizerEE.java:1141)
	at net.sf.saxon.expr.UserFunctionCall.optimize(UserFunctionCall.java:406)
	at com.saxonica.functions.hof.UserFunctionReference.optimize(UserFunctionReference.java:106)
	at net.sf.saxon.expr.Operand.optimize(Operand.java:206)
	at net.sf.saxon.expr.UnaryExpression.optimize(UnaryExpression.java:102)
	at net.sf.saxon.expr.Operand.optimize(Operand.java:206)
	at net.sf.saxon.expr.Expression.optimizeChildren(Expression.java:617)
	at net.sf.saxon.expr.Expression.optimize(Expression.java:594)
	at net.sf.saxon.expr.Operand.optimize(Operand.java:206)
	at net.sf.saxon.expr.Expression.optimizeChildren(Expression.java:617)
	at net.sf.saxon.expr.FunctionCall.optimize(FunctionCall.java:236)
	at net.sf.saxon.expr.SystemFunctionCall.optimize(SystemFunctionCall.java:238)
	at net.sf.saxon.expr.Operand.optimize(Operand.java:206)
	at net.sf.saxon.expr.Expression.optimizeChildren(Expression.java:617)
	at net.sf.saxon.expr.instruct.Block.optimize(Block.java:658)
	at net.sf.saxon.expr.Operand.optimize(Operand.java:206)
	at net.sf.saxon.expr.Expression.optimizeChildren(Expression.java:617)
	at net.sf.saxon.expr.FunctionCall.optimize(FunctionCall.java:236)
	at net.sf.saxon.expr.SystemFunctionCall.optimize(SystemFunctionCall.java:238)
	at net.sf.saxon.expr.Operand.optimize(Operand.java:206)
	at net.sf.saxon.expr.LetExpression.optimize(LetExpression.java:353)
	at net.sf.saxon.expr.UserFunctionCall.optimize(UserFunctionCall.java:409)
	at com.saxonica.functions.hof.UserFunctionReference.optimize(UserFunctionReference.java:106)
	at net.sf.saxon.expr.Operand.optimize(Operand.java:206)
	at net.sf.saxon.expr.UnaryExpression.optimize(UnaryExpression.java:102)
	at net.sf.saxon.expr.Operand.optimize(Operand.java:206)
	at net.sf.saxon.expr.Expression.optimizeChildren(Expression.java:617)
	at net.sf.saxon.expr.Expression.optimize(Expression.java:594)
	at net.sf.saxon.expr.Operand.optimize(Operand.java:206)
	at net.sf.saxon.expr.Expression.optimizeChildren(Expression.java:617)
	at net.sf.saxon.expr.FunctionCall.optimize(FunctionCall.java:236)
	at net.sf.saxon.expr.SystemFunctionCall.optimize(SystemFunctionCall.java:238)
	at net.sf.saxon.expr.Operand.optimize(Operand.java:206)
	at net.sf.saxon.expr.Expression.optimizeChildren(Expression.java:617)
	at net.sf.saxon.expr.instruct.Block.optimize(Block.java:658)
	at net.sf.saxon.expr.Operand.optimize(Operand.java:206)
	at net.sf.saxon.expr.Expression.optimizeChildren(Expression.java:617)
	at net.sf.saxon.expr.FunctionCall.optimize(FunctionCall.java:236)
	at net.sf.saxon.expr.SystemFunctionCall.optimize(SystemFunctionCall.java:238)
	at net.sf.saxon.expr.Operand.optimize(Operand.java:206)
	at net.sf.saxon.expr.LetExpression.optimize(LetExpression.java:353)
	at net.sf.saxon.expr.UserFunctionCall.optimize(UserFunctionCall.java:409)
	at com.saxonica.functions.hof.UserFunctionReference.optimize(UserFunctionReference.java:106)
...

Process finished with exit code 1
Actions #4

Updated by Debbie Lockett over 5 years ago

The transform works when the threshold is lowered, e.g. by setting this on the command-line: --thresholdForFunctionInlining:10.

Actions #5

Updated by Michael Kay over 5 years ago

I think that lowering the threshold prevents it happening for this particular case, simply because the function in question has a size above that threshold. That means that the bug is still there, it just needs a bigger function to trigger it.

The code currently has a defence mechanism to prevent infinite recursion during optimization, and it seems this defence mechanism isn't good enough. It prevents UserFunctionReference.optimize() doing anything if it is activated while it is already active; but the loop here is more subtle.

The code in UserFunctionReference.optimize() causes optimization of the body of the function that is referenced in the case where that function is an anonymous inline function: that's because we are essentially treating the inline function and the reference to it as a unit. I don't know the exact nature of the looping but it's not a simple recursion of the kind we defend against; it seems to involve a function being inlined, and the caller being optimized, which causes further inlining, ad infinitum. I think a suitable defence might be to limit the number of times any given function can be inlined - this is probably a useful thing to do anyway, since inlining a function thousands of times can cause memory bloat and also affects export files.

Actions #6

Updated by Michael Kay over 5 years ago

  • Category set to Internals
  • Status changed from In Progress to Resolved
  • Assignee set to Michael Kay

I have added a counter that checks how many times a function has been inlined; when it hits a threshold (currently 100), it is marked as non-inlineable. This fixes this particular test case.

Actions #7

Updated by Debbie Lockett over 5 years ago

Still seems to be some issue. Try the more complicated stylesheet taskBug2.xsl (again -it:ex1 -xsl:taskBug2.xsl). Without setting the threshold, it works eventually, but takes a looooong time.

Actions #8

Updated by Michael Kay about 5 years ago

There's clearly pathological compile time performance for taskBug2.xsl. The -explain trace shows a long sequence of function inlinings, like this:

OPT : At line 70 of file:/Users/mike/bugs/2019/2321-lockett/taskBug2.xsl
OPT : Moved function task:create-monadic-task inline: 
OPT : Expression after rewrite: let $apply-fn := (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_716157500#2))(?, $apply-fn) return (Q{http://www.w3.org/2005/xpath-functions/map}merge(Block(Q{http://www.w3.org/2005/xpath-functions/map}entry("apply", $apply-fn), Q{http://www.w3.org/2005/xpath-functions/map}entry("bind", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_984213526#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("then", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_942986815#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("fmap", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_1879034789#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("async", Q{http://ns.saxonica.com/anonymous-type}f_729864207#1($apply-fn))), net.sf.saxon.ma.map.DictionaryMap@5a01ccaa))
OPT : At line 70 of file:/Users/mike/bugs/2019/2321-lockett/taskBug2.xsl
OPT : Moved function task:create-monadic-task inline: 
OPT : Expression after rewrite: let $apply-fn := (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_716157500#2))(?, $apply-fn) return (Q{http://www.w3.org/2005/xpath-functions/map}merge(Block(Q{http://www.w3.org/2005/xpath-functions/map}entry("apply", $apply-fn), Q{http://www.w3.org/2005/xpath-functions/map}entry("bind", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_984213526#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("then", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_942986815#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("fmap", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_1879034789#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("async", Q{http://ns.saxonica.com/anonymous-type}f_729864207#1($apply-fn))), net.sf.saxon.ma.map.DictionaryMap@5a01ccaa))
OPT : At line 70 of file:/Users/mike/bugs/2019/2321-lockett/taskBug2.xsl
OPT : Moved function task:create-monadic-task inline: 
OPT : Expression after rewrite: let $apply-fn := (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_716157500#2))(?, $apply-fn) return (Q{http://www.w3.org/2005/xpath-functions/map}merge(Block(Q{http://www.w3.org/2005/xpath-functions/map}entry("apply", $apply-fn), Q{http://www.w3.org/2005/xpath-functions/map}entry("bind", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_984213526#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("then", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_942986815#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("fmap", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_1879034789#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("async", Q{http://ns.saxonica.com/anonymous-type}f_729864207#1($apply-fn))), net.sf.saxon.ma.map.DictionaryMap@5a01ccaa))
OPT : At line 70 of file:/Users/mike/bugs/2019/2321-lockett/taskBug2.xsl
OPT : Moved function task:create-monadic-task inline: 
OPT : Expression after rewrite: let $apply-fn := (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_716157500#2))(?, $apply-fn) return (Q{http://www.w3.org/2005/xpath-functions/map}merge(Block(Q{http://www.w3.org/2005/xpath-functions/map}entry("apply", $apply-fn), Q{http://www.w3.org/2005/xpath-functions/map}entry("bind", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_984213526#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("then", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_942986815#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("fmap", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_1879034789#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("async", Q{http://ns.saxonica.com/anonymous-type}f_729864207#1($apply-fn))), net.sf.saxon.ma.map.DictionaryMap@5a01ccaa))
OPT : At line 70 of file:/Users/mike/bugs/2019/2321-lockett/taskBug2.xsl
OPT : Moved function task:create-monadic-task inline: 
OPT : Expression after rewrite: let $apply-fn := (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_716157500#2))(?, $apply-fn) return (Q{http://www.w3.org/2005/xpath-functions/map}merge(Block(Q{http://www.w3.org/2005/xpath-functions/map}entry("apply", $apply-fn), Q{http://www.w3.org/2005/xpath-functions/map}entry("bind", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_984213526#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("then", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_942986815#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("fmap", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_1879034789#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("async", Q{http://ns.saxonica.com/anonymous-type}f_729864207#1($apply-fn))), net.sf.saxon.ma.map.DictionaryMap@5a01ccaa))
OPT : At line 70 of file:/Users/mike/bugs/2019/2321-lockett/taskBug2.xsl
OPT : Moved function task:create-monadic-task inline: 
OPT : Expression after rewrite: let $apply-fn := (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_716157500#2))(?, $apply-fn) return (Q{http://www.w3.org/2005/xpath-functions/map}merge(Block(Q{http://www.w3.org/2005/xpath-functions/map}entry("apply", $apply-fn), Q{http://www.w3.org/2005/xpath-functions/map}entry("bind", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_984213526#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("then", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_942986815#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("fmap", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_1879034789#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("async", Q{http://ns.saxonica.com/anonymous-type}f_729864207#1($apply-fn))), net.sf.saxon.ma.map.DictionaryMap@5a01ccaa))
OPT : At line 70 of file:/Users/mike/bugs/2019/2321-lockett/taskBug2.xsl
OPT : Moved function task:RUN-UNSAFE inline: 
OPT : Expression after rewrite: let $task := let $apply-fn := Q{http://ns.saxonica.com/anonymous-type}f_2030562336#1 return (Q{http://www.w3.org/2005/xpath-functions/map}merge(Block(Q{http://www.w3.org/2005/xpath-functions/map}entry("apply", $apply-fn), Q{http://www.w3.org/2005/xpath-functions/map}entry("bind", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_984213526#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("then", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_942986815#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("fmap", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_1879034789#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("async", Q{http://ns.saxonica.com/anonymous-type}f_729864207#1($apply-fn))))) return (tail(Q{http://saxon.sf.net/}apply(exactly-one((Q{http://saxon.sf.net/}apply($task, ["apply"])) treat as function(*)), SquareArrayConstructor($realworldEl))))
OPT : At line 92 of file:/Users/mike/bugs/2019/2321-lockett/taskBug2.xsl
OPT : Inlined references to $task
OPT : Expression after rewrite: tail(Q{http://saxon.sf.net/}apply(exactly-one(Q{http://saxon.sf.net/}apply(let $apply-fn := Q{http://ns.saxonica.com/anonymous-type}f_2030562336#1 return (Q{http://www.w3.org/2005/xpath-functions/map}merge(Block(Q{http://www.w3.org/2005/xpath-functions/map}entry("apply", $apply-fn), Q{http://www.w3.org/2005/xpath-functions/map}entry("bind", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_984213526#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("then", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_942986815#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("fmap", (fnCoercer(Q{http://ns.saxonica.com/anonymous-type}f_1879034789#2))(?, $apply-fn)), Q{http://www.w3.org/2005/xpath-functions/map}entry("async", Q{http://ns.saxonica.com/anonymous-type}f_729864207#1($apply-fn))))), ["apply"])), SquareArrayConstructor($realworldEl)))

and at this point it appears to hang.

Actions #9

Updated by Michael Kay about 5 years ago

The cut-off limit of 100 nested inlinings is taking effect, but there's something going wrong after it takes effect, and it's very hard to establish what, because by that time we've got 100 nested optimization operations on the stack all needing to be finished tidily. It feels like the cut-off mechanism is just too crude. We shouldn't be attempting inlining of a function as complex as this.

The reason it passes the complexity checks is that most of the complexity is within anonymous inline functions. These are treated as simple values; they don't contribute to the "expression size" calculation, and recursive calls within the anonymous functions are not considered; but they do add to the cost of the re-optimization that is performed after an inlining operation.

Actions #10

Updated by Michael Kay about 5 years ago

I have made two changes: (a) within OptimizerEE.tryInlineFunctionCall(), when optimizing the rewritten function call, we set a flag in the ExpressionVisitor that prevents further function inlining during this optimization phase; and (b) in UserFunctionCall.optimize(), we disable the optimization that occurs if the call has been inlined; this action is redundant because it has already been done as part of the inlining process.

With these changes, we avoid the deeply recursive repeated optimization, stopped only by the artificial max limit.

But it still hangs after the function inlining is complete. We now have a better chance of working out why it is hanging.

Actions #11

Updated by Michael Kay about 5 years ago

In UserFunctionReference.optimize, I have inserted a counter to show how many times the referenced anonymous function is being optimized. The counters get quite high, e.g. 3834, which is clearly excessive. However, they reach a limit before the process hangs/loops, so this is not the root cause of the problem.

After these counters stop rising, we're seeing a large (infinite?) number of calls on SystemFunctionCall.typeCheck(), especially on functions saxon:apply, map:entry, and map:merge. The JProfiler call graph suggests these are recursive.counter

Actions #12

Updated by Michael Kay about 5 years ago

I have solved the problem by adding counters to UserFunctionReference that limit the number of times a typeCheck() or optimize() call can cause the referenced function to be processed.

This change makes the change to ExpressionVisitor unnecessary, so I am reverting that. So the only changes are

(a) the counters in UserFunctionReference (b) UserFunctionCall no longer re-optimizes the function body after it is inlined.

Actions #13

Updated by Michael Kay about 5 years ago

  • Status changed from In Progress to Resolved
  • Applies to branch trunk added
  • Fix Committed on Branch trunk added
Actions #14

Updated by Debbie Lockett about 5 years ago

XSLT 3.0 test suite test added for regression testing: -s:higher-order-functions -t:higher-order-functions-075

(Test stylesheet is basically a renamed copy of taskBug2.xsl.)

Actions #15

Updated by O'Neil Delpratt almost 5 years ago

  • Status changed from Resolved to Closed

Please register to edit this issue

Also available in: Atom PDF