#7240: Stack trace truncated too much with indirect recursion
------------------------------+---------------------------------------------
Reporter: nomeata | Owner: nomeata
Type: bug | Status: new
Priority: normal | Component: Profiling
Version: 7.4.1 | Keywords:
Os: Unknown/Multiple | Architecture: Unknown/Multiple
Failure: None/Unknown | Testcase:
Blockedby: | Blocking:
Related: |
------------------------------+---------------------------------------------
Comment(by nomeata):
I started looking into this. Modifying ```checkLoop``` to only truncate if
there is a real loop at the top (e.g. ```<a,b,c,b>``` does not get
truncated yet, but ```<a,b,c,b,c>``` does get truncated to ```<a,b,c>```):
{{{
static CostCentreStack *
checkLoop (CostCentreStack *ccs, CostCentre *cc)
{
CostCentreStack *init_start, *top_seg, *bottom_seg;
init_start = ccs;
// Find previous instances of cc on the stack
while (init_start != EMPTY_STACK) {
if (init_start->cc == cc) {
// Compare the stack from here with the the top
top_seg = ccs;
bottom_seg = init_start->prevStack;
while (bottom_seg != EMPTY_STACK && top_seg->cc ==
bottom_seg->cc) {
// top_set != EMPTY_STACK as bottom_seg is definitely
below of
// it and not empty
top_seg = top_seg->prevStack;
bottom_seg = bottom_seg->prevStack;
}
if (top_seg == init_start) {
// We found that the segment above init_start equals the
segment below it,
// so we truncate to it.
return init_start;
}
// Otherwise, we try to find a larger repeating initial
segment.
}
init_start = init_start->prevStack;
}
return NULL;
}
}}}
The problem is that it does not play well with ```enterFunCCS```. I have
an example program with some recursion, where the untruncated version
would be ```<CAF,main,r,g,r,g,r,g,r,g,r,g,s,f>```, and one would expect,
with the above code, that this gets compactified to
```<CAF,main,r,g,s,f>```. But what I observe is
```<CAF,main,r,g,r,s,f>```.
Judging from the traces, this is due to ```enterFunCCC```. Where in the
untruncating code it is called with ccsfun=```<CAF,main,r,g,r,g>``` and
rCCCS=```<CAF,main,r,g,r>``` (yielding ```<CAF,main,r,g,r,g>```), now it
receives the shortened ccsfun=```<CAF,main,r,g>``` and adds the ```r```
back in, yielding ``<CAF,main,r,g,r>```. On this then ```s``` and ```f```
are pushed.
I wonder if such an effect can also occur with the current, simpler
truncating routine.
BTW, have you considered not truncating upon push at all, and then
simplifying the stack trace on display? This would not lose any
information and the ```enterFunCCS``` logic would work just fine. But
maybe the overhead is too large? In that case, would a RTS flag (“precise
recursive call stack”) be useful?
--
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/7240#comment:1>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler
_______________________________________________
Glasgow-haskell-bugs mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
