Which STL templates to use?

[pkirk25]

Hi all,

I am analysing a test file which documents auctions in World of
Warcraft. Its about 30MB and 500,000 lines of text.

Right now, I open the file and read it into a vector>string>. I chose
this becasue I thought it would be easier to navigate. Is there any
benefit to using ifstream instead?

When I analyse the data, I end up with quite a few tables which only
comprise of a key and a data item. For example "6661:0:0|Recipe:
Savory Deviate Delight" where "6661:0:0" is the key and "Recipe: Savory
Deviate Delight" is the value. The vector>stringis not likely to go
over 5000 in size, its entries are never changed though it can be added
to. Again I am storing these in vector>string>. Is that the best
option?

In choosing options, I am assuming that a user who palys World of
Warcraft has plenty of RAM and processing power. But if there are
benefts in terms of not using too much RAM or CPU power to an
alternative approach, I'd like to know as the main objective is to
learn C++ and the STL.

Thanks.

Patrick

[Gianni Mariani]

pkirk25 wrote:

Hi all,

I am analysing a test file which documents auctions in World of
Warcraft. Its about 30MB and 500,000 lines of text.

Right now, I open the file and read it into a vector>string>. I chose
this becasue I thought it would be easier to navigate. Is there any
benefit to using ifstream instead?

There are very efficient methods of processing files like this, however,
if you don't need to, worry about performance there is little need to
bother.

You need to know what your memory and cpu budget is and there has to be
a basis in fact behind it.

30MB is a little steep if you're talking about a background process that
runs all the time but if you're a gamer, it's a drop in the ocean ... so
it's not really a C++ question.

However, vector<stringis a fine way to do it, do not expect it to
break any performance expectations tho.

In the very few times I have had to have a really high performance
solution with file i/o, I has a special string type that would be set to
components of a memory mapped file. The 30MB of file does not get
copied to memory and only the components of the file that were changed
became in memory objects. Even faster is to use a binary file so that
the file in on disk has memory allocated in its own pool.

None of this stuff is C++ so we're way off topic for this NG.

The whole point is, before you worry about performance you need to make
sure you have a real problem to solve.

G