Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

core-clp: Rewrite wildcard matching method and add systematic unit tests (fixes #427). #428

Draft
wants to merge 11 commits into
base: main
Choose a base branch
from
267 changes: 155 additions & 112 deletions components/core/src/clp/string_utils/string_utils.cpp
Original file line number Diff line number Diff line change
@@ -1,72 +1,72 @@
#include "string_utils/string_utils.hpp"

#include <algorithm>
#include <charconv>
#include <cctype>
#include <cstring>

using std::string;
using std::string_view;

namespace clp::string_utils {
namespace {
/**
* Helper for ``wildcard_match_unsafe_case_sensitive`` to advance the pointer in
* tame to the next character which matches wild. This method should be inlined
* for performance.
* @param tame_current
* @param tame_bookmark
* @param tame_end
* @param wild_current
* @param wild_bookmark
* @return true on success, false if wild cannot match tame
* Helper for `wildcard_match_unsafe_case_sensitive` to advance `tame`'s iterator to the next
* character that matches the current character in `wild`, and to bookmark this character. If the
* current character in `wild` is escaped, `wild`'s iterator will also be advanced.
*
* NOTE:
* - This method expects that `tame_it` < `tame_end_it`
* - This method should be inlined for performance.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm not sure if this is enforced. afaik we are just suggesting the compiler to inline the method. But I'm also not sure whether we should use always_intline attribute since we may need to compile this file using none-gnu compilers.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It isn't enforced, but at the same time I don't know if forcing it to be inlined is necessary. The reason I said it should be inlined is because in past performance tests, the inline hint did make a difference. Nowadays though, it seems like gcc inlines it regardless of the hint.

*
* @param tame_end_it
* @param tame_it Returns `tame`'s updated iterator.
* @param tame_bookmark_it Returns `tame`'s updated bookmark.
* @param wild_it Returns `wild`'s updated iterator.
* @return true on success, false if `tame` cannot match `wild`.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
* @return true on success, false if `tame` cannot match `wild`.
* @return Whether `tame` can successfully match `wild`,

*/
inline bool advance_tame_to_next_match(
char const*& tame_current,
char const*& tame_bookmark,
char const* tame_end,
char const*& wild_current
string_view::const_iterator tame_end_it,
string_view::const_iterator& tame_it,
string_view::const_iterator& tame_bookmark_it,
string_view::const_iterator& wild_it
);

inline bool advance_tame_to_next_match(
char const*& tame_current,
char const*& tame_bookmark,
char const* tame_end,
char const*& wild_current
string_view::const_iterator tame_end_it,
string_view::const_iterator& tame_it,
string_view::const_iterator& tame_bookmark_it,
string_view::const_iterator& wild_it
) {
auto w = *wild_current;
auto w = *wild_it;
if ('?' != w) {
// No need to check for '*' since the caller ensures wild doesn't
// contain consecutive '*'
// No need to check for '*' since the caller ensures `wild` doesn't contain consecutive '*'

// Handle escaped characters
if ('\\' == w) {
++wild_current;
// This is safe without a bounds check since this the caller ensures
// there are no dangling escape characters
w = *wild_current;
// Safe without a bounds check
++wild_it;
w = *wild_it;
}

// Advance tame_current until it matches wild_current
// Advance `tame_it` until it matches `w`
while (true) {
if (tame_end == tame_current) {
// Wild group is longer than last group in tame, so can't match
// e.g. "*abc" doesn't match "zab"
return false;
}
auto t = *tame_current;
auto t = *tame_it;
if (t == w) {
break;
}
kirkrodrigues marked this conversation as resolved.
Show resolved Hide resolved
++tame_current;
++tame_it;
if (tame_end_it == tame_it) {
return false;
}
}
}

tame_bookmark = tame_current;
tame_bookmark_it = tame_it;

return true;
}
} // namespace

namespace clp::string_utils {
size_t find_first_of(
string const& haystack,
char const* needles,
Expand Down Expand Up @@ -182,114 +182,157 @@ bool wildcard_match_unsafe(string_view tame, string_view wild, bool case_sensiti

/**
* The algorithm basically works as follows:
* Given a wild string "*abc*def*ghi*", it can be broken into groups of
* characters delimited by one or more '*' characters. The goal of the algorithm
* is then to determine whether the tame string contains each of those groups in
* the same order.
* Given a wildcard string (a.k.a. "wild") like "abc*def*ghi*", it can be broken into groups of
* characters delimited by one or more '*' characters. The goal of the algorithm is then to
* determine whether the "tame" string contains each of those groups in the same order.
*
* Matching a group in `wild` against `tame` requires iteratively matching each character in `tame`
* against each character in the group, with the exception of the '?' wildcard and escaped
* characters ('*', '?', or '\'). When a mismatch occurs, there are two possibilities:
*
* 1. The mismatch occurs before the first '*' in `wild`, meaning that the entire wildcard match
* fails.
* 2. The mismatch occurs after a '*' in `wild`. This case requires additional handling explained
* below.
*
* Consider `tame` = "ccd", `wild` = "*cd". When we start matching `tame` against the first group
* in `wild`, the first 'c' will match, but the second 'c' won't match 'd'. In this case, we should
* restart the matching process from the second 'c'.
*
* To generalize this, we need to maintain bookmarks for both `tame` and `wild`. Whenever we have a
* mismatch, we should reset `wild` to its bookmark and `tame` to its bookmark + 1, and then try
* the match again. If we get to the end of `tame` without getting to the end of the group in
* `wild`, the entire wildcard match fails.
*
* Thus, the algorithm:
* 1. searches for the start of one of these groups in wild,
* 2. searches for a group in tame starting with the same character, and then
* 3. checks if the two match. If not, the search repeats with the next group in
* tame.
* NOTE:
* - Since the caller guarantees that there are no consecutive '*', we don't need to handle the
* case where a group in `wild` is empty.
* - Since the caller guarantees that every '\' is followed by a character, we can advance passed
* '\' without doing a subsequent bounds check.
* - The second part of this method could be rewritten in the following form:
*
* ```
* while(true) {
* if (false == advance_tame_to_next_match(...)) return false;
*
* while (true) {
* // Advance iterators
* // If we reached the end of `tame` before the end of `wild`, break
* // If we see a '*' in `wild`, break
* // If we see a mismatch, break
* }
* }
* ```
*
* However, this form is ~2% slower.
*/
bool wildcard_match_unsafe_case_sensitive(string_view tame, string_view wild) {
auto const tame_length = tame.length();
auto const wild_length = wild.length();
char const* tame_current = tame.data();
char const* wild_current = wild.data();
char const* tame_bookmark = nullptr;
char const* wild_bookmark = nullptr;
char const* tame_end = tame_current + tame_length;
char const* wild_end = wild_current + wild_length;
auto tame_it = tame.cbegin();
auto wild_it = wild.cbegin();
auto const tame_end_it = tame.cend();
auto const wild_end_it = wild.cend();
string_view::const_iterator tame_bookmark_it{};
string_view::const_iterator wild_bookmark_it{};

// Handle wild or tame being empty
if (0 == wild_length) {
return 0 == tame_length;
} else {
if (0 == tame_length) {
return "*" == wild;
// Handle `tame` or `wild` being empty
if (wild.empty()) {
return tame.empty();
}
if (tame.empty()) {
return "*" == wild;
}
Comment on lines +244 to +250
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why not moving the empty check to the beginning of the function?


// Match `tame` against `wild` against until we reach the first '*' in `wild` or they no longer
// match
while (true) {
auto w = *wild_it;
if ('*' == w) {
break;
}
if ('?' != w) {
// Handle escaped characters
if ('\\' == w) {
// Safe without a bounds check
++wild_it;
w = *wild_it;
}

// Handle a mismatch
if (w != *tame_it) {
return false;
}
}

++tame_it;
++wild_it;

// Handle boundary conditions
if (tame_end_it == tame_it) {
return (wild_end_it == wild_it) || (wild_end_it == wild_it + 1 && '*' == *wild_it);
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This line is repeated in the next for loop. How about we make a helper, sth like is_wild_reaching_end_or_trailing_star? (maybe we can discuss to come up with a better name)

} else if (wild_end_it == wild_it) {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do you think using a new if instead of else if would be more clear? Essentially these are two different cases to handle.

return tame_end_it == tame_it;
}
}

char w;
char t;
bool is_escaped = false;
// Find a match in `tame` for every group of characters between '*' in `wild`
while (true) {
w = *wild_current;
auto w = *wild_it;
if ('*' == w) {
++wild_current;
if (wild_end == wild_current) {
// Trailing '*' means everything remaining in tame will match
++wild_it;
if (wild_end_it == wild_it) {
// `wild` ending with '*' means that it'll match the rest of `tame`
return true;
}

// Set wild and tame bookmarks
wild_bookmark = wild_current;
// Set `tame` and `wild` bookmarks
wild_bookmark_it = wild_it;
if (false
== advance_tame_to_next_match(tame_current, tame_bookmark, tame_end, wild_current))
== advance_tame_to_next_match(tame_end_it, tame_it, tame_bookmark_it, wild_it))
{
return false;
}
} else {
} else if ('?' != w) {
// Handle escaped characters
if ('\\' == w) {
is_escaped = true;
++wild_current;
// This is safe without a bounds check since this the caller
// ensures there are no dangling escape characters
w = *wild_current;
// Safe without a bounds check
++wild_it;
w = *wild_it;
}

// Handle a mismatch
t = *tame_current;
if (!((false == is_escaped && '?' == w) || t == w)) {
if (nullptr == wild_bookmark) {
// No bookmark to return to
if (w != *tame_it) {
// Reset to bookmarks
tame_it = tame_bookmark_it + 1;
if (tame_end_it == tame_it) {
return false;
}

wild_current = wild_bookmark;
tame_current = tame_bookmark + 1;
wild_it = wild_bookmark_it;
if (false
== advance_tame_to_next_match(
tame_current,
tame_bookmark,
tame_end,
wild_current
))
== advance_tame_to_next_match(tame_end_it, tame_it, tame_bookmark_it, wild_it))
{
return false;
}
}
}

++tame_current;
++wild_current;
++tame_it;
++wild_it;

// Handle reaching the end of tame or wild
if (tame_end == tame_current) {
return (wild_end == wild_current
|| ('*' == *wild_current && (wild_current + 1) == wild_end));
} else {
if (wild_end == wild_current) {
if (nullptr == wild_bookmark) {
// No bookmark to return to
return false;
} else {
wild_current = wild_bookmark;
tame_current = tame_bookmark + 1;
if (false
== advance_tame_to_next_match(
tame_current,
tame_bookmark,
tame_end,
wild_current
))
{
return false;
}
}
// Handle boundary conditions
if (tame_end_it == tame_it) {
return (wild_end_it == wild_it) || (wild_end_it == wild_it + 1 && '*' == *wild_it);
} else if (wild_end_it == wild_it) {
if (tame_end_it == tame_it) {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

if tame_end_it == tame_it and wild_end_it == wild_it, we should already return in the above if right?

return true;
}

// Reset to bookmarks
tame_it = tame_bookmark_it + 1;
wild_it = wild_bookmark_it;
if (false
== advance_tame_to_next_match(tame_end_it, tame_it, tame_bookmark_it, wild_it))
{
return false;
Comment on lines +343 to +349
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Correct me if I'm wrong: if this branch is triggered, wild has already reached the end without consuming the entire tame. We should be handling the last group of tame after the last *. In this case, we only need to match the last n characters (determined by wild_it - wild_bookmark_it, and properly counting escape chars in between) in tame right? For example, if tame is "aaaaaaaa" and wild is "*a", we don't have to advance tame to match every single "a" but jump to match the last one

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You're right. Are you suggesting, in this case, that we should iterate backwards from the end of tame to see if it matches the last group in wild?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actually, iterating backwards is non-trivial because of escaped characters. If we see a ? in wild, we have to check the character before it to know if it's an escape character. But even if it is, we don't know if it's escaping the ? or it's preceded by an escape itself. So it's easier to always iterate forwards.

}
}
}
Expand Down
Loading
Loading