GeezerVison

# Search for tests related to mbcs codec
rg -n "mbcs" --type=rs --type=py -i

# Look for error handler validation patterns in the codebase
rg -n "errors.*strict|ignore|replace" --type=rs | head -30

# Check the broader context around codecs.rs to understand how other encode functions handle errors
rg -n "errors.*as_ref|errors.*unwrap_or" --type=rs -B2 -A2

# Check the codecs.rs file structure to see if there are other encode functions
cat -n crates/vm/src/stdlib/codecs.rs | head -100

# Check the entire mbcs_encode function and surrounding context
sed -n '237,310p' crates/vm/src/stdlib/codecs.rs

# Check if there are other encode functions to see how they handle errors
grep -n "fn.*encode" crates/vm/src/stdlib/codecs.rs

# Let's see what other encoding functions look like to understand error handling patterns
sed -n '1,150p' crates/vm/src/stdlib/codecs.rs

# Let's also check if there are any other error-related constants being used
rg -n "CP_ACP|WC_NO_BEST_FIT_CHARS" --type=rs -B5 -A5

# Look for any test files that test this function
find . -name "*.rs" -type f | xargs grep -l "mbcs" 2>/dev/null

# Check if there are Python stdlib tests we can reference
find . -name "*.py" -type f | xargs grep -l "mbcs" 2>/dev/null | head -10

# Let's check the entire file to understand the error handling context
wc -l crates/vm/src/stdlib/codecs.rs

# Check the full codecs.rs file for other encode functions and their error handling
sed -n '100,250p' crates/vm/src/stdlib/codecs.rs

# Let's check if there are any comments or documentation about error handling in this file
grep -n "error\|Error\|ERROR" crates/vm/src/stdlib/codecs.rs | head -20

# Let's understand what the "errors" parameter should control by checking Python documentation
# and looking at the actual Windows API behavior
# Check if this function validates the errors parameter properly
sed -n '237,250p' crates/vm/src/stdlib/codecs.rs

# Check what MbcsEncodeArgs structure looks like
rg -n "struct MbcsEncodeArgs" --type=rs -A10

# Let's look at the broader context of the file
sed -n '1,50p' crates/vm/src/stdlib/codecs.rs

# Find the MbcsEncodeArgs definition
grep -n "MbcsEncodeArgs" crates/vm/src/stdlib/codecs.rs

# Get the definition of MbcsEncodeArgs
sed -n '210,236p' crates/vm/src/stdlib/codecs.rs

# Let's check if there's a utf_8_encode or similar function to see how errors are handled there
sed -n '50,150p' crates/vm/src/stdlib/codecs.rs | grep -A20 "fn utf"

# Let's look for any utf_8_encode function in the file
rg -n "fn utf_8_encode" --type=rs -A30

# Let's understand the complete behavior by looking at lines around mbcs_encode
sed -n '230,330p' crates/vm/src/stdlib/codecs.rs

# Check commit history for mbcs_encode changes
git log --all -p --follow -S "mbcs_encode" -- crates/vm/src/stdlib/codecs.rs | head -100

# Check the git blame to understand the history
git blame -L237,310 crates/vm/src/stdlib/codecs.rs | head -20

In crates/vm/src/stdlib/codecs.rs around lines 237 to 308, the mbcs_encode
currently validates only "strict" and doesn't implement behavior for non-strict
modes: validate the errors argument against allowed modes
("strict","ignore","replace","xmlcharrefreplace","backslashreplace") and return
a ValueError for invalid names; always request WideCharToMultiByte to report
default-char usage (track used_default_char) and, after conversion, if
used_default_char!=0 handle unmappable characters according to the selected
mode: for "strict" raise a UnicodeEncodeError as now; for "ignore" remove the
bytes corresponding to unmappable input characters (skip them from the output
and adjust returned char count appropriately); for "replace" substitute
unmappable characters with a single replacement byte (e.g. b'?' or the system
best-fit char provided by the conversion) while preserving char length
semantics; for "backslashreplace" and "xmlcharrefreplace" post-process the
original UTF-8 input to replace each unmappable character with the corresponding
ASCII escape or XML character reference and then re-encode that sequence to
bytes; ensure WC_NO_BEST_FIT_CHARS is cleared or set appropriately per mode so
Windows does/doesn't do best-fit mapping, and update buffer
construction/truncation and returned char length to reflect any removals or
insertions.

In crates/vm/src/stdlib/codecs.rs around lines 517 to 598, oem_decode duplicates
mbcs_decode issues: it unconditionally recalculates the UTF‑16 buffer size
(making the earlier error-checking branch ineffective) and ignores the errors
parameter; refactor to follow the mbcs_decode fix by 1) compute required size
once using MB_ERR_INVALID_CHARS when errors == "strict", otherwise try
MB_ERR_INVALID_CHARS first and fall back to 0 and propagate the OS error only if
both calls return 0; 2) remove the redundant second size calculation and use the
returned size to allocate the buffer and perform the conversion; 3) use the
errors value (drop the leading underscore) to control strict vs replacement
behavior; and 4) extract a shared helper (e.g. windows_decode(data, code_page,
errors, codec_name, vm)) that contains the common logic and call it from
oem_decode with CP_OEMCP as the code_page to eliminate duplication.