x/blog: pprof example doesn't trigger reallocation

[vorishirne]

What version of Go are you using (go version)?

$ go version
go version go1.15.6 linux/amd64

Does this issue reproduce with the latest release?

Not related

What operating system and processor architecture are you using (go env)?

i5/ubuntu

What did you do?

In the golang blog of pprof, they demonstrate the performance of golang via a benchmarking code and then comparing with cpp counterpart. The blog tells about reallocartion of memory and its time complexities in multiple loops.

The blog => https://blog.golang.org/pprof

So, I builded the examples mentioned in the havlak repository.
But to find out, that the variable "size" is static throught the execution, to be precise, 252013.

What did you expect to see?

The constant value of size parameter, doesn't triggers the the memory re-allocation, measuring whose cost is the most focused part of the blog.

What did you see instead?

As the blog uses the havlok example as a memory critical task, It isn't actually.

[davecheney]

Did you include a link to the blog post?

[vorishirne]

Added in issue description, https://blog.golang.org/pprof

[vorishirne]

To add more context, printing here the diff of edited source code, and the output

diff --git a/../havlak/havlak6.go b/havlak6.go
index 39b2e3c..5530dd0 100644
--- a/../havlak/havlak6.go
+++ b/havlak6.go
@@ -291,6 +291,7 @@ func (f *LoopFinder) FindLoops(g *CFG, lsg *LoopGraph) {
        }
 
        // Step A: Initialize nodes, depth first numbering, mark dead nodes.
+       fmt.Println(size,cap(f.LoopBlock))
        if size <= cap(f.LoopBlock) {
                f.LoopBlock = f.LoopBlock[:size]
                f.DepthFirst = f.DepthFirst[:0]

For cpp code,

$ git diff ../havlak/havlak6.cc havlak6.cc
diff --git a/../havlak/havlak6.cc b/havlak6.cc
index 22afbdc..74ead0f 100644
--- a/../havlak/havlak6.cc
+++ b/havlak6.cc
@@ -269,7 +269,9 @@ void LoopFinder::FindLoops(CFG *g, LoopGraph *lsg) {
                return;
 
        // Step A: Initialize nodes, depth first numbering, mark dead nodes.
-       this->loopBlock.resize(size);
+       printf("%lu %d\n",loopBlock.size(),size);
+       if (0 < size){
+       this->loopBlock.resize(size);}
        this->depthFirst.reserve(size);
        this->depthFirst.clear();
        for (int i = 0; i < size; i++)

Program output :

$ ./xtime ./havlak6
252013 0
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
252013 252013
# of loops: 76000 (including 1 artificial root node)
2.77u 0.04s 2.53r 85940kB ./havlak6