Re: [PATCH v3 1/2] systemtap/tapsets.cxx: Adjusted for multiple static functions

[Hemant Kumar <hemant at linux dot vnet dot ibm dot com>]

On 04/08/2015 09:19 PM, Frank Ch. Eigler wrote:
> mjw wrote:
>
> > [...]
> > >     // TODO: Use a multimap in case there are multiple static
> > >     // functions with the same name?
> > >     map_by_addr.insert(make_pair(addr, fi));
> > But map_by_addr is already a multimap as introduced in commit 1c6b77
> > PR10327: resolve symbol aliases to dwarf functions by Josh. [...]
> > Since map_by_addr is using a multimap I was wondering if map_by_name
> > should also be a multimap instead of a map to a list? Do you happen to
> > know the advantages/disadvantages of the two datastructures?
> Indeed, the new structure should be a multimap too.  That means it'd
> be apprx. a binary tree of (key,value) pairs (with duplication
> allowed.  It would be a bit more canonical, and has advantages in
> cases where the list needs to be modified or traversed without
> copying.

Right.
But, when I tried that this with multimap, the problem I faced is we can 
insert two or more entries with same key and value, for e.g., if a 
multimap already has <a,b> as an entry, we can insert multiple <a,b> 
entries after that because they don't have duplication check which is 
not the case with simple maps.
But yeah, this can be done with multimap with a duplication check before 
insertion.

> > > +          list<func_info*> *fis = new list<func_info*>;
> > > +          fis = sym_table->lookup_symbol(function_str_val);
> > > [...]
> > Don't we need to delete the fis somewhere?
> lookup_symbol should just return the list by value and let C++ handle
> the memory management.  (The objects pointed to by the embedded
> func_info* pointers are a separate matter.)
>
> - FChE

--
Thanks,
Hemant Kumar