Advertisement

This article shows how to easily process UNIX-style directory entries as STL sequences.

The C++ community is gradually moving towards the STL paradigm—containers, iterators, function objects, algorithms and adaptors—which provides great advantages in terms of commonality of expression, reduction in developer effort, and greater robustness, maintainability and reuse. However, apart from the standard library and one or two exceptions (Boost [1], STLSoft [2]), there is relatively little STL-compliant code, which is, in part, due to the complexities involved in its development.

In this article I'll demonstrate how one kind of enumeration API—the UNIX opendir()/readdir() API—can be mapped into an STL-compliant sequence-like class providing Input Iterators. I examine the implementation of readdir_sequence, from UNIXSTL, the UNIX-specific sub-project of STLSoft ([2]).

Before getting into the details, consider the advantages of an STL-compliant approach. Let's say you need to load the names of the sub-directories of /home/matty/ into a vector of strings. Using the raw opendir() API, your code might look something like that shown in Listing 1.

It's not an enormous amount of code, to be sure, but it's still quite a bit for something so conceptually straightforward. Let's look at the readdir_sequence-based version, in Listing 2. It's clear in comparing the two code snippets that readdir_sequence represents a big win over the version using the opendir() API. There are several advantages:

1. The code is more succinct. That's always a popular improvement.
2. Closing the search, via closedir(), is automatically handled, via Resource Acquisition Is Initialisation, which improves robustness. Indeed, the first example is not exception-safe, since the std::string constructor and the std::vector<>::push_back() method may throw exceptions; the second version is exception-safe.
3. Elision of the dots directories—a very common requirement—is done automatically.
4. Directory and/or file selection is done via the appropriate constructor flag.
5. The effort in remembering the search directory and prefixing it to each entry name prior to stat-ing them is eliminated. All these aspects are handled internally by readdir_sequence::const_iterator.

The one flaw in the design is that the value_type is struct dirent const*, which means you have to explicitly enumerate the entries, rather than use algorithms or iterator-based constructors (see Sidebar).

Let's look now how it's implemented. Listing 3 shows the definition of the readdir_sequence class. It provides begin() and end() methods, which return iterators of member type const_iterator, since only non-mutating access to the entries is provided. Hence, readdir_sequence provides a read-only view on a directory.

It also provides an empty() method, which tests begin() against end(). I've deliberately omitted a size() member, since the size could only be obtained by conducting an enumeration over the range, which is a costly operation. Not providing this method is an unequivocal documentation of this fact [3]. If one really wants to calculate the range size, it can be done via std::distance().

For convenience to client code, a get_directory() method is provided which returns the search directory, having ensured, in the constructor, that it has a trailing path name separator ('/'). This means that if you need to express the returned values in absolute form, you can do so simply, as in:

dirNames.push_back(dir.get_directory() + (*b)->d_name);

The two parameters to the constructor are the directory to search and the flags, which control the search. I'll look at these shortly when I discuss the readdir_sequence::const_iterator class.

Since all the member variables of this class are fully-fledged value types, there is no need to proscribe or provide explicit implementations of either the copy constructor or the copy assignment operator. The compiler provided ones will work quite nicely and safely.

Another notable aspect to the class is that the type of the string member, m_directory, depends on whether the symbol PATH_MAX is defined [4]. It if is, then the operating environment has a fixed maximum path limit, and the class uses the STLSoft string basic_static_string, which has an internal array of the given size, so there's no memory allocation involved. If PATH_MAX is not defined, then the operating environment does not have a fixed maximum path limit, so the STLSoft string basic_simple_string is used instead [5]. It's a small efficiency, certainly, but I'm funny like that.

A further saving comes from the use of the m_scratch member. Since the directory is constant for the life of the sequence/iterator, and it's contents are not revealed directly outside the iterator instance, you are able to reuse it for constructing the full path names of the entries, in order to call stat() on them. The m_dirLen member remembers the original length of the directory alone, so it can be truncated to that length (which includes path name separator) ready for each entry name to be appended to it.

The opendir() API provides a two-step process to directory enumeration (in contrast to, say, Win32's FindFirstFile()), so the sequence class starts the enumeration—in its begin() method—and hands the DIR pointer over to the iterator to walk through the matched items.