Tra la la
A string in Solidity is length prefixed with its 256-bit (32 byte) length, which means that even short strings require 2 words (64 bytes) of storage.
In many cases, we deal with short strings, so instead of prefixing the string with its length, we can null-terminate it and fit it in a single word (32 bytes). Since we need only a single byte for the null termination, we can store strings up to 31 bytes long in a word.
Strings that are 31 bytes long may contain fewer than 31 characters, since UTF-8 requires multiple bytes to encode international characters.
Returns the decoded string represented by the
Bytes32 encoded data.
bytes32 string representation of text. If the length of text exceeds 31 bytes, it will throw an error.
Returns the UTF-8 bytes of text, optionally normalizing it using the UnicodeNormalizationForm form.
Returns the Array of codepoints of text, optionally normalized using the UnicodeNormalizationForm form.
This function correctly splits each user-perceived character into its codepoint, accounting for surrogate pairs. This should not be confused with
string.split(""), which destroys surrogate pairs, spliting between each UTF-16 codeunit instead.
There are several commonly used forms when normalizing UTF-8 data, which allow strings to be compared or hashed in a stable way.
Maintain the current normalization form.
The Composed Normalization Form. This form uses single codepoints which represent the fully composed character.
For example, the é is a single codepoint,
The Decomposed Normalization Form. This form uses multiple codepoints (when necessary) to compose a character.
For example, the é is made up of two codepoints,
"0x0065" (which is the letter
"0x0301" which is a special diacritic UTF-8 codepoint which indicates the previous character should have an acute accent.
The Composed Normalization Form with Canonical Equivalence. The Canonical representation folds characters which have the same syntactic representation but different semantic meaning.
For example, the Roman Numeral I, which has a UTF-8 codepoint
"0x2160", is folded into the capital letter I,
The Decomposed Normalization Form with Canonical Equivalence. See NFKC for more an example.
Only certain specified characters are folded in Canonical Equivalence, and thus it should not be considered a method to acheive any level of security from homoglyph attacks.
When converting a string to its codepoints, there is the possibility of invalid byte sequences. Since certain situations may need specific ways to handle UTF-8 errors, a custom error handling function can be used, which has the signature:
The reason is one of the UTF-8 Error Reasons, offset is the index into bytes where the error was first encountered, output is the list of codepoints already processed (and may be modified) and in certain Error Reasons, the badCodepoint indicates the currently computed codepoint, but which would be rejected because its value is invalid.
This function should return the number of bytes to skip past keeping in mind the value at offset will already be consumed.
UTF-8 Error Reasons
A byte was encountered which is invalid to begin a UTF-8 byte sequence with.
A UTF-8 sequence was begun, but did not have enough continuation bytes for the sequence. For this error the ofset is the index at which a continuation byte was expected.
The computed codepoint is outside the range for valid UTF-8 codepoints (i.e. the codepoint is greater than 0x10ffff). This reason will pass the computed badCountpoint into the custom error function.
Due to the way UTF-8 allows variable length byte sequences to be used, it is possible to have multiple representations of the same character, which means overlong sequences allow for a non-distinguished string to be formed, which can impact security as multiple strings that are otherwise equal can have different hashes.
Generally, overlong sequences are an attempt to circumvent some part of security, but in rare cases may be produced by lazy libraries or used to encode the null terminating character in a way that is safe to include in a
This reason will pass the computed badCountpoint into the custom error function, which is actually a valid codepoint, just one that was arrived at through unsafe methods.
The string does not have enough characters remaining for the length of this sequence.
This error is similar to BAD_PREFIX, since a continuation byte cannot begin a valid sequence, but many may wish to process this differently. However, most developers would want to trap this and perform the same operation as a BAD_PREFIX.
The computed codepoint represents a value reserved for UTF-16 surrogate pairs. This reason will pass the computed surrogate half badCountpoint into the custom error function.
There are already several functions available for the most common situations.
The will throw an error on any error with a UTF-8 sequence, including invalid prefix bytes, overlong sequences, UTF-16 surrogate pairs.
This will drop all invalid sequences (by consuming invalid prefix bytes and any following continuation bytes) from the final string as well as permit overlong sequences to be converted to their equivalent string.
This will replace all invalid sequences (by consuming invalid prefix bytes and any following continuation bytes) with the UTF-8 Replacement Character, (i.e. U+FFFD).