Due to my obsession with resource optimization, a while ago I thought: making the internet connection faster is certainly a good thing, but why not reduce what we send too?
This project aims to work on the second part of my thought: make the internet smaller.
The very basic idea is that, while HTML is human-readable, it is internet connection greedy, as a lot of unnecessary bytes are send over the network. To solve this problem it is sufficient to translate HTML to a more compact representation which operates on bytes instead of characters.
For instance the action attribute of a generic form shall be translated to
it's single byte equivalent 0x09, saving some precious bytes.
Of course it is required to add a little of semantic, so each (group of)
equivalent is precedeed by a special byte indicating it's type.
For instance we won't have just action but attribute, action, which
translates to 0x04 0x09.
The type is not strictly required as each element may have a different translation, but without it the required number of bytes to represent the whole HTML would explode. By means of a type instead, it is possible to subdivide the HTML into macro-groups (for instance tags, attributes, events, etc.) in which there will be equivalent overlapping.
Once a group ends, a special ending byte (0x00) is emitted, for instance the
group required checked will be translated to 0x04 0x5A 0x11 0x00.
Due to the fact that some developer may use non-standard HTML symbols, when the parser cannot find an equivalent for a given string, it is marked as text and copied as-is. By this way the algorithm is quite resilient.
A special case goes for multi-byte values, since most processor are little endian, this algorithm uses little endian as well.
Right now only attributes (with every possible type of value) have been encoded.
First of all, it is required to complete the whole HTML, then a similar algorithm for CSS will be implemented as well.
Everyone can contribute. Proofreading the documentation, bug-hunting, profiling, spread the word. Everything is good.
It like to put some special attenction to this points anyway:
- Test if output are correct
- Test on a big endian architecture
- Profile
- Implement SIMD