MDEV-10526: Add binary string support to bitwise operators

[kjarir]

Draft for review. Implements byte-by-byte binary string mode for all scalar bitwise operators. Aggregate function
support (BIT_AND/BIT_OR/BIT_XOR) to follow.

Tested:

• All 6 operators on VARBINARY columns
• INET6 subnet masking (primary use case)
• NULL handling
• Mismatched length error
• Integer mode backward compatibility
• _binary introducer triggers binary mode correctly

[gemini-code-assist]

Code Review

This pull request introduces support for binary-to-binary bitwise operations (AND, OR, XOR, NOT, and bitwise shifts) on binary strings by implementing new handler classes and adding corresponding error messages for mismatched operand sizes. The review feedback highlights several critical issues: a potential integer overflow/wrap-around in the left-shift index calculation on 32-bit systems, incorrect ASCII string conversion of numeric arguments in mixed-type bitwise operations, and potential undefined behavior when handling empty binary strings due to passing a null pointer to memset.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

The calculation size_t src_idx = i + byte_shift is susceptible to integer overflow/wrap-around on 32-bit systems if len is very large (e.g., close to size_t max). If i + byte_shift overflows, src_idx wraps around and can evaluate to a value less than len, leading to incorrect data copying from wrapped-around indices instead of zero-filling. To prevent this, perform the bounds check using subtraction (byte_shift < len - i) before computing src_idx.

        if (byte_shift < len - i)
        {
          size_t src_idx= i + byte_shift;

[gemini-code-assist]

In mixed-type bitwise operations (e.g., binary_string & numeric), calling val_str on the numeric argument converts it to its decimal string representation (e.g., 255 becomes the 3-byte ASCII string "255" / 0x323535), rather than its binary representation (e.g., 0xFF). This causes mismatched length errors (e.g., VARBINARY(1) & 255 compares length 1 with length 3) or incorrect bitwise operations if the lengths happen to match. To align with standard bitwise behavior (and MySQL 8.0 compatibility), numeric arguments should be converted to a binary string representation of their integer value, padded or truncated to match the length of the other binary string operand.

[gemini-code-assist]

When the input binary string a is empty (len == 0), calling to->realloc(0) can return a null pointer or do nothing. Subsequently, passing a null pointer to memset (e.g., memset(out_ptr, 0, len)) is technically undefined behavior in C/C++, even if the length is 0. Adding an early exit for len == 0 avoids this potential undefined behavior and improves efficiency by bypassing unnecessary allocation and loop overhead.

    size_t len= a->length();
    if (len == 0)
    { 
      to->length(0);
      to->set_charset(&my_charset_bin);
      item->null_value= false;
      return to;
    }

    if (to->realloc(len))

[grooverdan]

Should be off the main branch as a new feature.

from @abarkov

"Would be nice to add tests that uuid, inet6, inet4, geometry are not allowed for bit operations. Later we can probably implement bit operations for uuid, inet6, inet4. But to avoid compatibility problems we need to make sure they return error now."

[grooverdan]

from @abarkov
"check that max_length is calculated correctly in all functions"

This will show up in mtr --cursor-protocol on test that have truncated test result output on fields.

[CLAassistant]

All committers have signed the CLA.