Buffer#toString throw on unsupported encodings

[SheetJSDev]

Fixes #2417

New behavior: if argument is not undefined/unspecified or "utf8", throw an error.

Note: null is not a supported encoding in NodeJS

[SheetJSDev]

The core logic has been factored out since both Buffer.isEncoding and Buffer#toString have to check the encoding.

For the case insensitive matching, raw bit-ops are preferred since the strcmp variants (stricmp / strcasecmp) and the ctypes.h utilities (tolower / toupper) are not currently used.

Both Buffer#toString and Buffer.isEncoding tests have been refreshed.

[svaarala]

Thanks for the revisions 👍

[svaarala]

This loop works incorrectly if i reaches the maximum encoding length without finding a NUL terminator in the input: buf will remain unterminated and it's used for a strcmp() later.

Inputs like utf8\u0000bar are (technically) handled incorrectly, as the loop will terminate on the NUL.

ORing 0x20 on all input chars also results in some incorrect inputs accepted, e.g. "utf\u000d8" will be accepted because 0x0d | 0x20 = 0x2d which is '-'.

So overall I don't think approach is easy to make work.

[svaarala]

I wonder if it'd produce less code to just short circuit compare the input, e.g.:

if ((p[0] | 0x20) == 'u' && (p[1] | 0x20) == 't' && (p[2] | 0x20) == 'f') {
    if (p[3] == '8' && encoding_len == 4) {
        /* utf8 */
    } else if (p[3] == '-' && p[4] == '8' && encoding_len == 5) {
        /* utf8 */
    }
}
/* invalid */

[svaarala]

I'm merging the clang-format indent changes to master and to v2-maintenance. Once that's done, the easiest approach is probably to clang-format your changed files first, and then rebase against master/v2-maintenance. This should keep the diff minimal and avoid conflicts.

[svaarala]

Ok, clang-format changes are now in master and v2-maintenance. To reindent code:

$ make docker-image-clang-format
$ make clang-format-source