TextView

Synopsis

#include <ts/TextView.h>`
class TextView

This class acts as a view in to memory allocated / owned elsewhere. It is in effect a pointer and should be treated as such (e.g. care must be taken to avoid dangling references by knowing where the memory really is). The purpose is to provide string manipulation that is fast, efficient, and non-modifying, particularly when temporary "copies" are needed.

Description

TextView is a subclass of std::string_view and has all of those methods. In addition it provides a number of ancillary methods of common string manipulation methods.

A TextView should be treated as an enhanced character pointer that both a location and a size. This is when makes it possible to pass sub strings around without having to make copies or allocation additional memory. This comes at the cost of keeping track of the actual owner of the string memory and making sure the TextView does not outlive the memory owner, just as with a normal pointer type. Internal for Traffic Server any place that passes a char * and a size is an excellent candidate for using a TextView as it is more convenient and no more risky than the existing arguments.

In deciding between std::string_view and TextView remember that these easily and cheaply cross convert. In general if the string is treated as a block of data, std::string_view is better. If the contents of the string are to be examined / parsed non-uniformly then TextView is better. For example, if the string is used simply as a key or a hash source, use std::string_view. Or, if the string may contain substrings of interests such as key / value pairs, then use a TextView.

TextView provides a variety of methods for manipulating the view as a string. These are provided as families of overloads differentiated by how characters are compared. There are four flavors.

  • Direct, a pointer to the target character.

  • Comparison, an explicit character value to compare.

  • Set, a set of characters (described by a TextView) which are compared, any one of which matches.

  • Predicate, a function that takes a single character argument and returns a bool to indicate a match.

If the latter three are inadequate the first, the direct pointer, can be used after finding the appropriate character through some other mechanism.

The increment operator for TextView shrinks the view by one character from the front which allows stepping through the view in normal way, although the string view itself should be the loop condition, not a dereference of it.

TextView v;
size_t hash = 0;
for ( ; v ; ++v) hash = hash * 13 + * v;

Because the view acts as a container of characters, this can be done non-destructively.

TextView v;
size_t hash = 0;
for (char c : v) hash = hash * 13 + c;

Views are cheap to construct therefore making a copy to use destructively is very inexpensive.

MemSpan provides a find method that searches for a matching value. The type of this value can be anything that is fixed sized and supports the equality operator. The view is treated as an array of the type and searched sequentially for a matching value. The value type is treated as having no identity and cheap to copy, in the manner of a integral type.

Parsing with TextView

A primary use of TextView is to do field oriented parsing. It is easy and fast to split strings in to fields without modifying the original data. For example, assume that value contains a null terminated string which is possibly several tokens separated by commas.

#include <ctype.h>
parse_token(const char* value) {
  TextView v(value); // construct assuming null terminated string.
  while (v) {
    TextView token(v.extractPrefix(','));
    token.trim(&isspace);
    if (token) {
      // process token
    }
  }
}

If value was bob  ,dave, sam then token would be successively bob, dave, sam. After sam was extracted value would be empty and the loop would exit. token can be empty in the case of adjacent delimiters or a trailing delimiter. Note that no memory allocation at all is done because each view is a pointer in to value and there is no need to put nul characters in the source string meaning no need to duplicate it to prevent permanent changes.

What if the tokens were key / value pairs, of the form key=value? This is can be done as in the following example.

#include <ctype.h>
parse_token(const char* source) {
  TextView in(source); // construct assuming null terminated string.
  while (in) {
    TextView value(in.extractPrefix(','));
    TextView key(value.trim(&isspace).splitPrefix('=').rtrim(&isspace));
    if (key) {
      // it's a key=value token with key and value set appropriately.
      value.ltrim(&isspace); // clip potential space after '='.
    } else {
      // it's just a single token which is in value.
    }
  }
}

Nested delimiters are handled by further splitting in a recursive way which, because the original string is never modified, is straight forward.

History

The first attempt at this functionality was in the TSConfig library in the ts::Buffer and ts::ConstBuffer classes. Originally intended just as raw memory views, ts::ConstBuffer in particular was repeated enhanced to provide better support for strings. The header was eventually moved from lib/tsconfig to lib/ts and was used in various parts of the Traffic Server core.

There was then a proposal to make these classes available to plugin writers as they proved handy in the core. A suggested alternative was Boost.StringRef which provides a similar functionality using std::string as the base of the pre-allocated memory. A version of the header was ported to Traffic Server (by stripping all the Boost support and cross includes) but in use proved to provide little of the functionality available in ts::ConstBuffer. If extensive reworking was required in any case, it seemed better to start from scratch and build just what was useful in the Traffic Server context.

The next step was the TextView class which turned out reasonably well. It was then suggested that more support for raw memory (as opposed to memory presumed to contain printable ASCII data) would be useful. An attempt was made to do this but the differences in arguments, subtle method differences, and return types made that infeasible. Instead MemSpan was split off to provide a void* oriented view. String specific methods were stripped out and a few non-character based methods added.