Re: [Boost-users] [proto, fusion] Segfault in for_each

Subject: Re: [Boost-users] [proto, fusion] Segfault in for_each
From: Eric Niebler (eric_at_[hidden])
Date: 2010-05-19 14:06:09

On 5/19/2010 10:27 AM, Eric Niebler wrote:
> On 5/19/2010 10:01 AM, David Greene wrote:
>> On Thursday 13 May 2010 23:16:31 David A. Greene wrote:
>>> On Monday 10 May 2010 21:15:42 Eric Niebler wrote:
>>>>> I've been wrestling with this for days.  Any ideas on how to debug
>>>>> this?
>>>>
>>>> Nope. But it'd probably help if you sent around a self-contained repro.
>>>
>>> Ok, here's a smallish testcase.  gdb shows the same problem: an invalid
>>> proto expression reference.
>>
>> Hey Eric,
>>
>> No rush to take a look at this, but I wanted to make sure you got the
>> testcase since I haven't seen a reply.
> 
> Sorry for not responding sooner. I got it. This came in during the melee
> of BoostCon. And now I'm down with a cold. I hope to look at it soon.

OK, I had a quick look. Basically, you're playing fast and loose with
Fusion views and temporaries. When you call fusion::transform, it
doesn't eagerly generate a new fusion sequence. It returns a /view/ that
holds a reference to the input sequence and the map function. In
transformList and all the functions it calls, you're passing arguments
by value, and then returning a fusion::transform_view object that is
holding a reference to the (now deceased) argument. I've changed the
code to pass-by-reference to side-step the lifetime issues, and it seems
to work now. I make no promises that there aren't other lifetime issues
in this code. You need to think carefully about temporaries and lifetime
management when using proto and fusion.

I also fixed your grammar for you. Be sure to draw a distinction between
grammar rules and expressions. terminal<int> is a grammar rule.
terminal<int>::type is an expression. An expression can be used as a
grammar (it matches itself), but a grammar is not an expression.

HTH,

-- 
Eric Niebler
BoostPro Computing
http://www.boostpro.com