Das Problem mit German Strings | Hacker News Reader

Das Problem mit German Strings

polarsignals.com

79 pointsasubiotto23 comments

> Because it is difficult to assume what the best encoding will be for any given workload, database systems should dynamically choose encodings based on storage and workload characteristics.

It would be better just to take the storage requirement on the chin and not add a gratuitous variation in encoding which will bite you on the ass somehow (or someone else).

As much as possible, pick one way of doing one thing. Your stuff already has thousands of things to do. Each time you do something in two or more ways, you add combinations between that and surrounding things being done in two or more ways.

The combinatorial explosion problem is nicely solved by defining good interfaces. C++ gives you iterators and algorithms that work on iterators. Clojure has sequence interfaces and functions that work on all sequence types.

That just improves the organization of the program; it doesn't get rid of the increased risks of doing the same thing in N ways that could be pined down to one.

Please elaborate. What are the risks of doing the same thing in N ways, other than code organization issues leading to duplicate or messy code?

Do this thing in 3 ways, do that one in 4, do another one in 2 and you have 3x4x2 = 24 combinations which are entirely gratuitous compared to the 1 combination that exists if all three things are done one way each.

Oh, you don't have to test the combinations because the code is bug free, is that the argument? Which is because of some good organization?

Those things are nicely isolated so 3 + 4 + 2 unit tests, and we are done?

> Because each element requires at least a 16 byte representation, both tiny and repeated short strings use more memory than they otherwise would.

In a wider view, that depends. If one is using a general-purpose heap for string storage and a 64-bit instruction set architecture, the heap is often aligning and padding out allocations to such multiples already.

Well as long as you know the difference betwen lowercase ß and uppercase ẞ (introduced in 2008) everything is probably just gonna be fine.

So... why are they called Getman strings?

https://datafusion.apache.org/blog/2024/09/13/string-view-ge...

> The concept of inlined strings with prefixes (called “German Strings” by Andy Pavlo, in homage to TUM, where the Umbra paper that describes them originated) has been used in many recent database systems (Velox, Polars, DuckDB, CedarDB, etc.) and was introduced to Arrow as a new StringViewArray[^3] type. Arrow’s original StringArray is very memory efficient but less effective for certain operations. StringViewArray accelerates string-intensive operations via prefix inlining and a more flexible and compact string representation.

Seems to be nothing more than they were invented at a German university. I spent quite some time thinking it had something to do with German’s sometimes-SOV word order.

> I spent quite some time thinking it had something to do with German’s sometimes-SOV word order.

If you refer to subclauses in the German language: here the rule is rather "the finite verb is at the end of the subclause".

It also applies to infitives and participles and the verb in nominalized noun-verb compounds. So the rule is closer to "the verb is at the end of its grammatical unit, except for the finite verb in a main clause, which appears in second position." https://en.wikipedia.org/wiki/V2_word_order

I think this is also called V2 word order.

V2 word order (finite verb comes second) is what is used in main clauses.

This general string format style has been invented many times over the decades. Unfortunately, we seem to need to relearn the tradeoffs each time.

Here is the paper in question:

Umbra: A Disk-Based System with In-Memory Performance

https://db.in.tum.de/~freitag/papers/p29-neumann-cidr20.pdf

Section 3.1 covers string handling.

This article (also linked from tfa) explains German strings in more detail.

https://cedardb.com/blog/german_strings

my tl;dr: after reading the article:

- two 64-bits words representation

- fixed, 32 bits length

- short strings (<12 bytes) are stored in-place

- long strings store a 4 byte prefix in-place + pointer to the rest

- two bits are used as flags in the pointer to further optimize some use-cases

Seems like they missed an opportunity to have a 8 byte version for strings that fit in the 4 byte prefix.

They aren't. They're called German style strings. People just like to clickbait and prey on curiosity of techies.

did the hacker news title editor change the "mit" to "MIT"?

Seems like it. Changed it back!

Oops, sorry.

Haha, is that automated or was someone trying to be helpful?

It's automated. And of course it's usually right, but the wrong cases stand out like sore thumbs.

Reply on news.ycombinator.com