Add serialized capacity to Cache_Redis for objects

[chonhan]

Updated with Tcholakov's suggestion.
Verified on Both Mac and Ubuntu environments and run correctly with my database result.

[narfbg]

• Remove this constant
• Hardcode the string to '_ci_redis_serialized:'

On second thought ... it should be a constant, just give it a better name.

[ivantcholakov]

I have chosen this ugly suffix in order it not to be hit by a developer by accident. I am fine with '_ci_redis_serialized:' too. As I can see below, by this string you mean a prefix.

[narfbg]

Overall, I see this as something rather hacky ... isn't there a way to separately mark the keys of serialized values? Like creating a list containing all serialized vars? Such a list would be easily read and cached during the library initialization and way easier to work with.

[narfbg]

Also, I got lost between all those PRs, just commit to one of them.

[ivantcholakov]

Could I answer these questions?

[narfbg]

Sure, if you know the answers.

[ivantcholakov]

I did not comment the obvious things "Add empty line.", etc. They are to be done.

You figured out a third approach, this is interesting. Let me enumerate the possible ways:

1. Direct prefixing the serialized value - I abandoned this one, a serialized value may be 10Mb long for example, I don't want to add/check prefix there.
2. Prefixed keys for serialized values - the current implementation. It has no the disadvantage of 1., it is yet ready and it is tested. I wrote the test and applied driver modifications simultaneously, behavior is visible. Yes code might look "hackish", but it is explainable.
3. Index of serialized values - your approach. If it is to pass the same test it will look "hackish" I afraid. There is one obstacle to be solved explicitly, the time factor - there are values that expire. At the moment I have no idea how to keep the index synchronized. The idea is interesting, but maybe it is quite challenging, at least for me.

Well, if you like the second approach, you may merge chonhan's contribution as it is now. I promise to polish it after that, according to your comments.

[narfbg]

Why would it look hackish? All you'll need to do is fetch that list during the library initialization and add to it in save(). Anywhere else you'd just make an in_array() check before (un)serializing ...
The time factor is not an issue. You only care about those lookups after you've fetched the value, so if the item has expired, you'd just return FALSE prior to that. Not that you can't track time of course, it's just not necessary.

[ivantcholakov]

OK, I will try to implement the third approach, slowly, without rush. It makes sense.

[ivantcholakov]

Just in case, before the next try I have polished the second idea by implementing the remarks by @narfbg. Here is the gist: https://gist.github.com/ivantcholakov/e1d889793e469315ec89

[narfbg]

Implemented in #3186.