mirror of
https://github.com/vlang/v.git
synced 2023-08-10 21:13:21 +03:00
64 lines
1.3 KiB
Go
64 lines
1.3 KiB
Go
|
module chunked
|
||
|
|
||
|
import strings
|
||
|
|
||
|
// See: https://en.wikipedia.org/wiki/Chunked_transfer_encoding
|
||
|
///////////////////////////////////////////////////////////////
|
||
|
// The chunk size is transferred as a hexadecimal number
|
||
|
// followed by \r\n as a line separator,
|
||
|
// followed by a chunk of data of the given size.
|
||
|
// The end is marked with a chunk with size 0.
|
||
|
|
||
|
struct ChunkScanner {
|
||
|
mut:
|
||
|
pos int
|
||
|
text string
|
||
|
}
|
||
|
|
||
|
fn (s mut ChunkScanner) read_chunk_size() int {
|
||
|
mut n := 0
|
||
|
for {
|
||
|
if s.pos >= s.text.len { break }
|
||
|
c := s.text[s.pos]
|
||
|
if !c.is_hex_digit() { break }
|
||
|
n = n << 4
|
||
|
n += int(unhex(c))
|
||
|
s.pos++
|
||
|
}
|
||
|
return n
|
||
|
}
|
||
|
|
||
|
fn unhex(c byte) byte {
|
||
|
if `0` <= c && c <= `9` { return c - `0` }
|
||
|
else if `a` <= c && c <= `f` { return c - `a` + 10 }
|
||
|
else if `A` <= c && c <= `F` { return c - `A` + 10 }
|
||
|
return 0
|
||
|
}
|
||
|
|
||
|
fn (s mut ChunkScanner) skip_crlf() {
|
||
|
s.pos += 2
|
||
|
}
|
||
|
|
||
|
fn (s mut ChunkScanner) read_chunk(chunksize int) string {
|
||
|
startpos := s.pos
|
||
|
s.pos += chunksize
|
||
|
return s.text.substr(startpos, s.pos)
|
||
|
}
|
||
|
|
||
|
pub fn decode(text string) string {
|
||
|
mut sb := strings.new_builder(100)
|
||
|
mut cscanner := ChunkScanner {
|
||
|
pos: 0
|
||
|
text: text
|
||
|
}
|
||
|
for {
|
||
|
csize := cscanner.read_chunk_size()
|
||
|
if 0 == csize { break }
|
||
|
cscanner.skip_crlf()
|
||
|
sb.write( cscanner.read_chunk(csize) )
|
||
|
cscanner.skip_crlf()
|
||
|
}
|
||
|
cscanner.skip_crlf()
|
||
|
return sb.str()
|
||
|
}
|