Embed hash value in hash type entry

[zuiderkwast]

For a hash type key represented as a hash table, embed the field and value in one allocation when they together fit in the size of one cache line. For larger fields and values, another layout is used where the value is a separately allocated sds string.

Implementation

The hashTypeEntry pointer is changed to be the field sds, i.e. a pointer to the embedded field content, which is located after the sds header in memory. We encode the entry layout in the field sds field for unused space, so sdsavail(entry) gives the entry encoding.

Entry with embedded field and value, used when it fits in a cache line:

+--------------+----------------+--------------+
| field        | 1 byte         | value        |
| hdr "foo" \0 | value-hdr-size | hdr "bar" \0 |
+------^-------+----------------+--------------+
       |
       |
     entry pointer = field sds

Entry with value-pointer and embedded field, used for larger field-value pairs:

+-------+--------------+
| value | field        |
| ptr   | hdr "foo" \0 |
+-------+------^-------+
               |
               |
            entry pointer = field sds

This implementation does not require the fields nor values to be of a specific SDS encoding, but it does convert the field from sds5 to sds8 in the second entry layout (the value-pointer embedded-field), so we can encode the type. An embedded sds5 field implies the first entry layout.

Fixes #1551.

[codecov]

Codecov Report

Attention: Patch coverage is 88.33333% with 14 lines in your changes missing coverage. Please review.

Project coverage is 71.03%. Comparing base (7fc958d) to head (e40ccfd).

Files with missing lines	Patch %	Lines
src/t_hash.c	88.13%	14 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##           unstable    #1579      +/-   ##
============================================
+ Coverage     70.80%   71.03%   +0.23%     
============================================
  Files           121      121              
  Lines         65132    65226      +94     
============================================
+ Hits          46118    46336     +218     
+ Misses        19014    18890     -124     

Files with missing lines	Coverage Δ
src/object.c	82.04% <100.00%> (ø)
src/sds.c	86.87% <100.00%> (+0.02%)	⬆️
src/server.h	100.00% <ø> (ø)
src/t_hash.c	94.72% <88.13%> (-1.41%)	⬇️

... and 10 files with indirect coverage changes

[ranshid]

May I suggest something?
We can add another type of sdscopytobuffer which forces a specific type and can also be used by sdscopytobuffer:

/* This method copies the sds `s` into `buf` which is the target character buffer. */
size_t sdscopytobufferwithtype(unsigned char *buf, size_t buf_len, char type, const sds s, uint8_t *hdr_size) {
    assert((s[-1] & SDS_TYPE_MASK) <= type);
    size_t header_len = sdsHdrSize(type);
    size_t str_len = sdslen(s);
    size_t alloc_size = header_len + str_len + 1;
    if (hdr_size) *hdr_size = sdsHdrSize(type);
    if (buf == NULL) {
        return alloc_size;
    }
    assert(buf_len >= alloc_size);
    sds embed_sds = (sds)(buf + header_len);
    embed_sds[-1] = type & SDS_TYPE_MASK;
    sdssetlen(embed_sds, str_len);
    sdssetalloc(embed_sds, alloc_size);
    memcpy(embed_sds, s, str_len + 1);
    return alloc_size;
}

[ranshid]

Sorry for turning into Scrooge McDuck.
I actually thought about this the opposite way around. I think we can FORCE non sds5 fields only when the value IS embedded.
So the question of: canUseEmbeddedValueEntry can take into account the new field sds header length and the value sds size.
This may also allow us to use the alloc of embedded fields to hold the value header size, so when the field sds alloc is 0 I actually know this is a ENTRY_ENC_PTR_VALUE and in case it is not 0 I know it is ENTRY_ENC_EMB_VALUE.