json-validator/include/jvalidate/uri.h

#pragma once

#include <cstdlib>
#include <ostream>
#include <string>
#include <string_view>

#include <jvalidate/compat/compare.h> // IWYU pragma: keep

namespace jvalidate {
/**
 * @brief A subsection of the Uniform Resource Identifier (URI) syntax as per
 * RFC 3986 (https://datatracker.ietf.org/doc/html/rfc3986).
 *
 * This URI structure supports file paths (either as relative paths or as
 * file:// URIs), URNs (because they are covered by the JSON-Schema test suite),
 * and HTTP/S urls (assigning the consumer the responsibility of handling e.g.
 * params).
 * Additionally - this URI implementation does not support fragment parts. This
 * is because in the context of a JSON schema - a fragment part is treated as
 * either an Anchor, or as a JSON-Pointer.
 *
 * Because of these limitations, it is safe to treat this URI type as a tuple of
 * (scheme, resource) without the connecting "://" or ":" field.
 */
class URI {
private:
  std::string uri_;
  size_t scheme_{0};
  size_t resource_{0};

public:
  URI() = default;

  explicit URI(std::string_view uri) : uri_(uri) {
    // Special handling for some parsing situations where we know that an object
    // is a URI (and thus don't need to call Reference(text).uri()) - but that
    // URI may or may not contain a trailing hash (fragment indicator). This is the
    // case with the "$schema" field, for example. For any given draft, the schema
    // writer can start with "http://" OR "https://", and might end with a "#".
    if (not uri_.empty() && uri_.back() == '#') {
      uri_.pop_back();
    }

    // Locate file://, http://, and https:// schemes
    if (size_t pos = uri_.find("://"); pos != std::string::npos) {
      scheme_ = pos;
      resource_ = pos + 3;
    } else if (uri_.starts_with("urn:")) {
      // Note that we skip over the first colon, because the format of a URN
      // token is "urn:format:data" - and therefore we want the scheme to be
      // "urn:format", with the resource element to be "data".
      pos = uri_.find(':', 4);
      scheme_ = pos;
      resource_ = scheme_ + 1;
    }
  }

  URI parent() const { return URI(std::string_view(uri_).substr(0, uri_.rfind('/'))); }
  URI root() const { return URI(std::string_view(uri_).substr(0, uri_.find('/', resource_))); }

  /**
   * @brief "Concatenate" two URIs together. Most of the logic behind this
   * is done in {@see ReferenceManager}, rather than this class/function.
   * Included below are some example use-cases:
   *
   * "file://A/B/C" / "D.json" => "file:/A/B/C/D.json"
   * "http://example.com/foo" / "bar/baz.json" =>
   *    "http://example.com/foo/bar/baz.json"
   * "http://example.com/foo" / "/bar/baz.json" =>
   *    "http://example.com/bar/baz.json" (notice that we lost foo)
   *
   * Note that example 3 is not achieved through this function, but through code
   * in ReferenceManager that says something like:
   * @code{.cpp}
   * if (not relative.is_relative()) {
   *    uri = uri.root() / relative;
   * }
   * @endcode
   *
   * @param relative The "relative" URI to append to this one. Even though I say
   * relative, this URI may start with a leading "/", as long as it is rootless.
   * In that case, this URI is expected to be an HTTP/S URI - and we are going
   * to replace everything after the hostname with the contents of relative.
   */
  URI operator/(URI const & relative) const {
    if (relative.uri_.starts_with("/")) {
      return URI(uri_ + relative.uri_);
    }
    if (relative.uri_.starts_with("./")) {
      return URI(uri_ + relative.uri_.substr(1));
    }
    return URI(uri_ + "/" + relative.uri_);
  }

  /**
   * @brief Synonym for "does not have a scheme", used for short-circuiting
   * relative URI handling.
   */
  bool is_rootless() const { return scheme_ == 0; }
  /**
   * @brief Even if a URI does not have a scheme, it could still be non-relative
   * item, such as the URI "/dev/null" - which unambiguously refers to to root
   * directory (in a *nix type filesystem) - as opposed to "dev/null", which
   * could mean different resources in different parent contexts.
   *
   * Given that the "$id" that we set acts similar to `cd` in a shell, knowing
   * this let's us know if we're looking at "an entirely separate location", or
   * a "child/sibling location".
   */
  bool is_relative() const { return is_rootless() && uri_[resource_] != '/'; }

  std::string_view scheme() const { return std::string_view(uri_).substr(0, scheme_); }
  std::string_view resource() const { return std::string_view(uri_).substr(resource_); }

  explicit operator std::string const &() const { return uri_; }
  char const * c_str() const { return uri_.c_str(); }
  bool empty() const { return uri_.empty(); }

  friend std::ostream & operator<<(std::ostream & os, URI const & self) { return os << self.uri_; }

  auto operator<=>(URI const & lhs) const = default;
};
}