Implement legacy semantics for string escaping behavior #85
Conversation
| } | ||
| q, _ := jsontext.AppendQuote(nil, b) // cannot fail since b is valid UTF-8 | ||
| return enc.WriteValue(q) |
There was a problem hiding this comment.
Previously, the escaping for EscapeForHTML and EscapeForJS was deferred to the natural behavior of enc.WriteValue.
However, to replicate existing v1 behavior, we escape earlier in L210 by passing in mo.Flags, which will have the escape flags specified.
| b, _ := AppendUnquote(nil, src[:n]) | ||
| dst, _ = AppendQuote(dst, string(b), flags) | ||
| dst, _ = AppendQuote(dst, b, flags) |
There was a problem hiding this comment.
Unnecessary casting of []byte to string removed now that AppendQuote is a generic function operating on either type.
| @@ -162,6 +162,8 @@ func (v *Value) reformat(canonical, multiline bool, prefix, indent string) error | |||
| eo.Flags.Set(jsonflags.AllowInvalidUTF8 | 1) | |||
| eo.Flags.Set(jsonflags.AllowDuplicateNames | 1) | |||
| eo.Flags.Set(jsonflags.PreserveRawStrings | 1) | |||
| eo.Flags.Set(jsonflags.EscapeForHTML | 0) // ensure strings are preserved | |||
| eo.Flags.Set(jsonflags.EscapeForJS | 0) // ensure strings are preserved | |||
There was a problem hiding this comment.
These are 0 by default anyways, but explicitly set to 0 for clarity.
Yep... I haven't been running any of the benchmarks lately and I'm somewhat afraid to find out the result... |
For doubly escaped strings, the v1 behavior was to apply escaping for EscapeForHTML and EscapeForJS on the first pass, rather than on the second pass. This does lead to unnecessarily longer outputs since escaping on the first pass means that the newly included '\\' characters have to be escaped again in the second pass. Doubly escaped strings are already an esoteric feature of v1, so we should preserve the semantic. Also, when encoding the raw output of MarshalJSON, the v1 behavior was to preserve any pre-existing escape sequences, while still escaping any characters that might need escaping per EscapeForHTML and EscapeForJS. Replicate this behavior.
For doubly escaped strings, the v1 behavior was to apply escaping for EscapeForHTML and EscapeForJS on the first pass, rather than on the second pass. This does lead to unnecessarily longer outputs since escaping on the first pass means that the newly included '\' characters have to be escaped again in the second pass. Doubly escaped strings are already an esoteric feature of v1, so we should preserve the semantic.
Also, when encoding the raw output of MarshalJSON, the v1 behavior was to preserve any pre-existing escape sequences, while still escaping any characters that might need escaping per EscapeForHTML and EscapeForJS. Replicate this behavior.