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

gh-110289: C API: Add PyUnicode_EqualToUTF8() function #110297

Merged

Conversation

serhiy-storchaka
Copy link
Member

@serhiy-storchaka serhiy-storchaka commented Oct 3, 2023

Copy link
Member

@vstinner vstinner left a comment

Choose a reason for hiding this comment

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

PyUnicode_EqualToString() inline UTF-8 encoder is hard for review for me right now, I would feel more comfortable with tests, especially on corner cases:

  • string not encoded to UTF-8
  • Evil surrogate characters

Doc/c-api/unicode.rst Show resolved Hide resolved
assert(str);
if (PyUnicode_IS_ASCII(unicode)) {
size_t len = (size_t)PyUnicode_GET_LENGTH(unicode);
return strlen(str) == len &&
Copy link
Member

Choose a reason for hiding this comment

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

I would prefer to test the length first, to make the code more readable.

Like:

if (strlen(str) == len) {
    return 1;
}
return memcmp(...);

Same below.

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 is the same in _PyUnicode_EqualToASCIIString().

How

if (!a) {
    return 0;
}
return b;

is more readable than simple return a && b;? It is what the && operator for.

Copy link
Member

Choose a reason for hiding this comment

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

is more readable than simple return a && b;?

For me, it's easier to reason about a single test per line when I review code.

Keep a && b if you prefer.

Copy link
Contributor

Choose a reason for hiding this comment

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

The readability problem as I see it, is that your && use has side effects; it is not a pure logic expression.

Copy link
Member Author

Choose a reason for hiding this comment

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

For me, it's easier to reason about a single test per line when I review code.

Fortunately, every condition here is already on a separate line.

The readability problem as I see it, is that your && use has side effects; it is not a pure logic expression.

It is how && works in C. There is a lot of code like arg != NULL and PyDict_Check(arg) && PyDict_GET_SIZE(arg) > count. I do not think rewriting it in three ifs with gotos can improve readability.

@vstinner
Copy link
Member

vstinner commented Oct 3, 2023

Suggestion for a different function name to avoid any confusion... and make it shorter: PyUnicode_EqualToUTF8().

@serhiy-storchaka
Copy link
Member Author

I considered two variants: PyUnicode_EqualToUTF8String() and PyUnicode_EqualToString().

@serhiy-storchaka serhiy-storchaka changed the title gh-110289: C API: Add PyUnicode_EqualToString() function gh-110289: C API: Add PyUnicode_EqualToUTF8() function Oct 3, 2023
Doc/c-api/unicode.rst Outdated Show resolved Hide resolved
Comment on lines 1401 to 1402
Compare a Unicode object with a UTF-8 encoded C string and return true
if they are equal and false otherwise.
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
Compare a Unicode object with a UTF-8 encoded C string and return true
if they are equal and false otherwise.
Compare a Unicode object with a UTF-8 encoded C string and return non-zero
if they are equal or 0 otherwise.

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 looks to me, that "return true" is more often used than "return non-zero". In this case it is more accurate, because it always returns 1, not other non-zero value. Perhaps other functions which return non-zero was a macro that returned not 1 (something like (arg->flags & FLAG))?

Comment on lines 1005 to 1006
a :c:expr:`const char*` UTF-8 encoded bytes string and return true if they
are equal or false otherwise.
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
a :c:expr:`const char*` UTF-8 encoded bytes string and return true if they
are equal or false otherwise.
a :c:expr:`const char*` UTF-8 encoded bytes string and return non-zero if they
are equal or 0 otherwise.

Lib/test/test_capi/test_unicode.py Outdated Show resolved Hide resolved
Lib/test/test_capi/test_unicode.py Outdated Show resolved Hide resolved
Lib/test/test_capi/test_unicode.py Outdated Show resolved Hide resolved
Lib/test/test_capi/test_unicode.py Outdated Show resolved Hide resolved
Doc/c-api/unicode.rst Show resolved Hide resolved
Doc/c-api/unicode.rst Show resolved Hide resolved
}
else if (ch < 0x800) {
if ((0xc0 | (ch >> 6)) != (unsigned char)*str++ ||
(0x80 | (ch & 0x3f)) != (unsigned char)*str++)
Copy link
Member

Choose a reason for hiding this comment

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

unsigned char byte1 = (0xc0 | (ch >> 6));
unsigned char byte2 = (0x80 | (ch & 0x3f));
if (str[0] != byte1 || str[1] != byte2) return 0;

And declare a str variable as unsigned char* once to avoid casting str at each byte comparison.

Copy link
Member Author

Choose a reason for hiding this comment

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

If the first comparison fails, you do not need to calculate the second byte. The code looks more compact and uniform in the way it is written right now. All expressions I copied from the UTF-8 encoder which I wrote 11 years ago, so no need to recheck them. Casting to unsigned char is not a large burden, but if you prefer, I can introduce a new unsigned char* variable.

@serhiy-storchaka serhiy-storchaka marked this pull request as ready for review October 4, 2023 08:35
@serhiy-storchaka serhiy-storchaka requested review from a team and encukou as code owners October 4, 2023 08:35
Doc/c-api/unicode.rst Show resolved Hide resolved
@@ -1396,6 +1396,18 @@ They all return ``NULL`` or ``-1`` if an exception occurs.
:c:func:`PyErr_Occurred` to check for errors.


.. c:function:: int PyUnicode_EqualToUTF8(PyObject *unicode, const char *string)

Compare a Unicode object with a UTF-8 encoded C string and return true (``1``)
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
Compare a Unicode object with a UTF-8 encoded C string and return true (``1``)
Compare a Unicode object with a UTF-8 encoded or ASCII encoding C string and return true (``1``)

Copy link
Member Author

Choose a reason for hiding this comment

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

Maybe "ASCII encoded"?

assert(str);
if (PyUnicode_IS_ASCII(unicode)) {
size_t len = (size_t)PyUnicode_GET_LENGTH(unicode);
return strlen(str) == len &&
Copy link
Member

Choose a reason for hiding this comment

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

is more readable than simple return a && b;?

For me, it's easier to reason about a single test per line when I review code.

Keep a && b if you prefer.

Lib/test/test_capi/test_unicode.py Outdated Show resolved Hide resolved
Modules/_testcapi/unicode.c Outdated Show resolved Hide resolved
assert(str);
if (PyUnicode_IS_ASCII(unicode)) {
size_t len = (size_t)PyUnicode_GET_LENGTH(unicode);
return strlen(str) == len &&
Copy link
Contributor

Choose a reason for hiding this comment

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

The readability problem as I see it, is that your && use has side effects; it is not a pure logic expression.

@serhiy-storchaka
Copy link
Member Author

I tried to rewrite the code in more vertically sparse way:

int
PyUnicode_EqualToUTF8(PyObject *unicode, const char *str)
{
    assert(_PyUnicode_CHECK(unicode));
    assert(str);

    if (PyUnicode_IS_ASCII(unicode)) {
        size_t len = (size_t)PyUnicode_GET_LENGTH(unicode);
        if (strlen(str) != len) {
            return 0;
        }
        if (memcmp(PyUnicode_1BYTE_DATA(unicode), str, len) != 0) {
            return 0;
        }
        return 1;
    }
    if (PyUnicode_UTF8(unicode) != NULL) {
        size_t len = (size_t)PyUnicode_UTF8_LENGTH(unicode);
        if (strlen(str) != len) {
            return 0;
        }
        if (memcmp(PyUnicode_UTF8(unicode), str, len) != 0) {
            return 0;
        }
        return 1;
    }

    const unsigned char *s = (const unsigned char *)str;
    Py_ssize_t len = PyUnicode_GET_LENGTH(unicode);
    int kind = PyUnicode_KIND(unicode);
    const void *data = PyUnicode_DATA(unicode);
    /* Compare Unicode string and UTF-8 string */
    for (Py_ssize_t i = 0; i < len; i++) {
        Py_UCS4 ch = PyUnicode_READ(kind, data, i);
        if (ch == 0) {
            return 0;
        }
        else if (ch < 0x80) {
            if (s[0] != ch) {
                return 0;
            }
            s += 1;
        }
        else if (ch < 0x800) {
            if (s[0] != (0xc0 | (ch >> 6))) {
                return 0;
            }
            if (s[1] != (0x80 | (ch & 0x3f))) {
                return 0;
            }
            s += 2;
        }
        else if (ch < 0x10000) {
            if (Py_UNICODE_IS_SURROGATE(ch)) {
                return 0;
            }
            if (s[0] != (0xe0 | (ch >> 12))) {
                return 0;
            }
            if (s[1] != (0x80 | ((ch >> 6) & 0x3f))) {
                return 0;
            }
            if (s[2] != (0x80 | (ch & 0x3f))) {
                return 0;
            }
            s += 3;
        }
        else {
            assert(ch <= MAX_UNICODE);
            if (s[0] != (0xf0 | (ch >> 18))) {
                return 0;
            }
            if (s[1] != (0x80 | ((ch >> 12) & 0x3f))) {
                return 0;
            }
            if (s[2] != (0x80 | ((ch >> 6) & 0x3f))) {
                return 0;
            }
            if (s[3] != (0x80 | (ch & 0x3f))) {
                return 0;
            }
            s += 4;
        }
    }
    return *s == 0;
}
and it causes dizziness and eye pain in me. It is physically difficult for me to read it.

Doc/c-api/unicode.rst Show resolved Hide resolved
# CRASHES equaltoutf8(b'abc', b'abc')
# CRASHES equaltoutf8([], b'abc')
# CRASHES equaltoutf8(NULL, b'abc')
# CRASHES equaltoutf8('abc') # NULL
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
# CRASHES equaltoutf8('abc') # NULL
# CRASHES equaltoutf8('abc', NULL)

Copy link
Member Author

Choose a reason for hiding this comment

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

No, it does not work so.

NULL is defined as None, and equaltoutf8('abc', None) is a TypeError.

If equaltoutf8() is called with only one argument, it sets the second argument for PyUnicode_EqualToUTF8() to NULL, so we can test it and ensure that it indeed crashes. It is a common approach used in other tests in this file for const char * argument. Some functions do not crash, but raise exception or return successfully for NULL, but this function simply crashes in debug build.

Copy link
Member

Choose a reason for hiding this comment

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

Oh ok, I thought that they were just pseudo-code as comments. Sure, you can leave # NULL if you prefer.

Copy link
Member Author

Choose a reason for hiding this comment

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

Hmm, I copied this pattern from the test for PyUnicode_CompareWithASCIIString() which was one of the first written tests. In newer tests I use "z#" which allows to pass None for NULL. Or perhaps I changed this everywhere except the test for PyUnicode_CompareWithASCIIString(). So perhaps I can change this too.

Lib/test/test_capi/test_unicode.py Show resolved Hide resolved
Modules/_testcapi/unicode.c Outdated Show resolved Hide resolved
Lib/test/test_capi/test_unicode.py Outdated Show resolved Hide resolved
Doc/whatsnew/3.13.rst Outdated Show resolved Hide resolved
Doc/c-api/unicode.rst Outdated Show resolved Hide resolved
Doc/c-api/unicode.rst Outdated Show resolved Hide resolved
@vstinner
Copy link
Member

vstinner commented Oct 4, 2023

I prepared a PR to add this function to Python 2.7-3.12 in the pythoncapi-compat project: python/pythoncapi-compat#78

I chose to write a simple implementation:

        utf8 = PyUnicode_AsUTF8AndSize(unicode, &len);
        if (utf8 == NULL) {
            // Memory allocation failure. The API cannot report error,
            // so clear the exception and return 0.
            PyErr_Clear();
            return 0;
        }

It's tempting to ask you to modify the API to return -1 on error, but on the other side I hate APIs with simple tasks like "compare two strings" which can fail :-( Most people simply... don't check for errors.

So well, I like the propose API, function which cannot fail.

@serhiy-storchaka
Copy link
Member Author

Oh, other features of this implementation:

  • It can be called when an error is set and preserves it.
  • It does not use heap, so it can be used when MemoryError has been raised.

Copy link
Member

@vstinner vstinner left a comment

Choose a reason for hiding this comment

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

LGTM. I just left a few more minor comments.

Include/unicodeobject.h Outdated Show resolved Hide resolved
# CRASHES equaltoutf8(b'abc', b'abc')
# CRASHES equaltoutf8([], b'abc')
# CRASHES equaltoutf8(NULL, b'abc')
# CRASHES equaltoutf8('abc') # NULL
Copy link
Member

Choose a reason for hiding this comment

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

Oh ok, I thought that they were just pseudo-code as comments. Sure, you can leave # NULL if you prefer.

Modules/_testcapi/unicode.c Outdated Show resolved Hide resolved
@encukou
Copy link
Member

encukou commented Oct 5, 2023

IMO we ned a general strategy around dealing with strings. Let's not solve just for PyUnicode_Equal, but design something that we'll also use for, say, dict and attribute lookup.

Having two functions, for for both zero-terminated str and for separate length argument, sounds good to me. And we also want a third one that takes PyUnicode. (Yes, in this case we have it already).

Which of those should be in what kind of C-API? Which should be in stable ABI, which can just be inline functions? What should the naming conventions be? Is the char* const? What's the thread safety strory?
Please delay merging until after the sprint -- I hope to come up with a proposal for how to answer questions like that, consistently.

@vstinner
Copy link
Member

vstinner commented Oct 5, 2023

Which of those should be in what kind of C-API?

The 3 flavors should be exposed as regular function calls.

@vstinner
Copy link
Member

vstinner commented Oct 5, 2023

This is 2023 and null-encoded C strings are definitely not a good idea for new C APIs.

Would you mind to elaborate why/how using null terminated C string became a bad thing in 2023?

@pitrou
Copy link
Member

pitrou commented Oct 5, 2023

Would you mind to elaborate why/how using null terminated C string became a bad thing in 2023?

"Became a bad thing in 2023" is your interpretation. It has always been a design mistake, but it becomes even more glaring when interoperating with other languages which made the correct decision (that is, strings in those languages store their size explicitly).

In the distant times when the CPython C API was only called from C software, expecting null-terminated strings was fine, but it's not anymore.

@vstinner
Copy link
Member

vstinner commented Oct 5, 2023

In the distant times when the CPython C API was only called from C software, expecting null-terminated strings was fine, but it's not anymore.

I don't thin that it's worth to argue. We should just add an API without size, and an API with a size. That's all.

The API without size is at least needed to upgrade all users of _PyUnicode_Equal() and _PyUnicode_EqualToASCIIId(), removed in Python 3.13.

@pitrou
Copy link
Member

pitrou commented Oct 5, 2023

I don't thin that it's worth to argue.

Ah... I'm reassured, thank you.

@vstinner
Copy link
Member

vstinner commented Oct 5, 2023

Oh, apparently this PR is now discussed at https://discuss.python.org/t/new-pyunicode-equaltoutf8-function/35377

s += 4;
}
}
return *s == 0;
Copy link
Contributor

Choose a reason for hiding this comment

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

I suppose that if we return true at this point then we know that str is the utf8 representation of unicode, does it make sense to copy the contents into unicode->utf8 so that future operations can fast-path without needing to encode again?

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 needs a separate research and discussion. The disadvantage is that it increases the consumed memory size, also it consumes some CPU time, so the benefit will be only if the UTF-8 cache is used in future.

If the idea turned out to be good, it can simply be implemented in the future.

Copy link
Contributor

@davidhewitt davidhewitt Oct 5, 2023

Choose a reason for hiding this comment

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

Makes total sense. I guess this also sits in an awkward place where it's likely that the user is best suited to know whether or not they want the utf-8 cache populated, but it's also an implementation detail that we don't really want to expose to users. For now I'll just mark this comment as resolved. Edit I can't, probably lack permissions I guess.

Copy link
Member

Choose a reason for hiding this comment

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

PyUnicode_EqualToUTF8() doesn't raise exception and cannot fail. Trying to allocate memory should not raise memory, but it sounds like a non-trivial side effect.

Worst case: 1 GB string, you call PyUnicode_EqualToUTF8() and suddenly, Python allocates 1 GB more. I would be surprised by this behavior.

Copy link
Member

Choose a reason for hiding this comment

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

Maybe it's worth it to add a comment explaining why we don't cache the UTF-8 encoded string.

Doc/c-api/unicode.rst Outdated Show resolved Hide resolved
@@ -1396,18 +1396,28 @@ They all return ``NULL`` or ``-1`` if an exception occurs.
:c:func:`PyErr_Occurred` to check for errors.


.. c:function:: int PyUnicode_EqualToUTF8(PyObject *unicode, const char *string)
.. c:function:: int PyUnicode_EqualToUTF8AndSize(PyObject *unicode, const char *string, Py_ssize_t size)
Copy link
Member

Choose a reason for hiding this comment

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

What do you think about renaming string to utf8_str? The utf8_ would be another way to document that it's expected to be encoded to UTF-8 and also it's easier (for me) to distinguish that the second argument is a bytes string, since string name is quite generic.

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 is a part of the bigger issue. See #62897.

Doc/c-api/unicode.rst Show resolved Hide resolved
Lib/test/test_capi/test_unicode.py Outdated Show resolved Hide resolved
Objects/unicodeobject.c Outdated Show resolved Hide resolved
Copy link
Member

@vstinner vstinner left a comment

Choose a reason for hiding this comment

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

LGTM the updated PR which now also adds PyUnicode_EqualToUTF8AndSize(). You just have to fix the merge conflict.

So far, I didn't see any real blocker issue in the healthy discussion.

Lib/test/test_capi/test_unicode.py Outdated Show resolved Hide resolved
Lib/test/test_capi/test_unicode.py Outdated Show resolved Hide resolved
@erlend-aasland erlend-aasland dismissed their stale review October 10, 2023 20:46

Offending docstrings were removed; dismissing my request for changes

@serhiy-storchaka
Copy link
Member Author

@pitrou, does it look good to you now?

[function.PyUnicode_EqualToUTF8]
added = '3.13'
[function.PyUnicode_EqualToUTF8AndSize]
added = '3.13'
Copy link
Member

Choose a reason for hiding this comment

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

Unrelated question, but is there a plan to generate this file from Doc/data/stable_abi.dat or the reverse?

Copy link
Member Author

Choose a reason for hiding this comment

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

I think that Doc/data/stable_abi.dat is generated from Misc/stable_abi.toml.

Copy link
Member

Choose a reason for hiding this comment

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

Ah, ok, thank you!

@serhiy-storchaka serhiy-storchaka merged commit eb50cd3 into python:main Oct 11, 2023
27 checks passed
@serhiy-storchaka serhiy-storchaka deleted the capi-PyUnicode_EqualToString branch October 11, 2023 13:42
@vstinner
Copy link
Member

I added these functions to pythoncapi-compat: python/pythoncapi-compat@99ab0d3

Glyphack pushed a commit to Glyphack/cpython that referenced this pull request Sep 2, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

7 participants