public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
* [Bug libstdc++/13537] New: std::deque Is Not Conformant
@ 2004-01-01  4:16 stl at caltech dot edu
  2004-01-01  4:49 ` [Bug libstdc++/13537] " pinskia at gcc dot gnu dot org
                   ` (7 more replies)
  0 siblings, 8 replies; 9+ messages in thread
From: stl at caltech dot edu @ 2004-01-01  4:16 UTC (permalink / raw)
  To: gcc-bugs

23.2.1.3.3 says, "Inserting a single element either at the beginning or end of 
a deque always takes constant time".

(Aside: Does the Standard mean true constant time there?

23.2.4.1 says, "A vector is a kind of sequence that supports random access 
iterators.  In addition, it supports (amortized) constant time insert and 
erase operations at the end".  So the Standard does recognize the difference 
between amortized constant time and true constant time.

Of course, 23.1.1.12 says, "The operations in Table 68 are provided only for 
the containers for which they take constant time", and then goes on to list 
push_back for vector.

So the Standard sometimes says "constant time" in places where it really 
means "amortized constant time".

But note the exact phrasing of 23.2.1.3.3. "always takes constant time" 
clearly rules out the possibility that it is referring to amortized constant 
time.)

*** Thus, I believe that libstdc++ is not conformant. ***

It seems (and I am not familiar with deque's implementation at all) that 
a "map" of pointers is maintained in the deque, and insertions at the front or 
back can cause the map to be reallocated.  This takes O(N) time, even if the 
map's size is N / 4096 or whatever.

Looking at the source for 3.3.2:

stl_deque.h: push_back() can call _M_push_back_aux().
deque.tcc: _M_push_back_aux() calls _M_reserve_map_at_back() with no arguments.
stl_deque.h: _M_reserve_map_at_back() takes __nodes_to_add which is by default 
1.
    It can thus call _M_reallocate_map(1, false).
deque.tcc: _M_reallocate_map() gets called with __nodes_to_add = 1 and 
__add_at_front = false.
    It can call copy(_M_start._M_node, _M_finish._M_node + 1, __new_nstart).
    This looks like it is O(N).

Admittedly this bug is rather pedantic, but situations could be envisioned 
where amortized constant time insertions cause total protonic reversal.

(This bug was found by ryanm@microsoft.com ; we discussed it, I verified that 
he wasn't going crazy, and I agreed to submit a PR.)

-- 
           Summary: std::deque Is Not Conformant
           Product: gcc
           Version: 3.3.2
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: libstdc++
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: stl at caltech dot edu
                CC: gcc-bugs at gcc dot gnu dot org,ryanm at microsoft dot
                    com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13537


^ permalink raw reply	[flat|nested] 9+ messages in thread
end of thread, other threads:[~2004-01-29  1:35 UTC | newest]

Thread overview: 9+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2004-01-01  4:16 [Bug libstdc++/13537] New: std::deque Is Not Conformant stl at caltech dot edu
2004-01-01  4:49 ` [Bug libstdc++/13537] " pinskia at gcc dot gnu dot org
2004-01-01  5:33 ` stl at caltech dot edu
2004-01-11 10:08 ` paolo at gcc dot gnu dot org
2004-01-11 10:11 ` paolo at gcc dot gnu dot org
2004-01-11 10:16 ` paolo at gcc dot gnu dot org
2004-01-11 16:45 ` paolo at gcc dot gnu dot org
2004-01-28 22:27 ` skirby at acm dot org
2004-01-29  1:35 ` stl at caltech dot edu

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).