Strings Just Got Faster

https://inside.java/2025/05/01/strings-just-got-faster/

by Tomte • 1 year ago

246 181 1 year ago

inside.java

gavinray3• 1 year ago

This post makes mention of a new JEP I hadn't heard of before: "Stable Values"

https://openjdk.org/jeps/502

https://cr.openjdk.org/~pminborg/stable-values2/api/java.bas...

I don't understand the functional difference between the suggested StableValue and Records, or Value Classes.

They define a StableValue as:

  > "A stable value is a holder of contents that can be set at most once."

Records were defined as:

  > "...  classes that act as transparent carriers for immutable data. Records can be thought of as nominal tuples."

And Value Objects/Classes as:

  > "... value objects, class instances that have only final fields and lack object identity."

Both Records and Value Objects are immutable, and hence can only have their contents set upon creation or static initalization.

sagacity3• 1 year ago

Handwavy explanation: A stable value is used as a static constant, with the difference being that you can initialize it at runtime. Once initialized it is treated as fully constant by the JVM. It's similar to something like lateinit in Kotlin, except on the JVM level.

Records are also immutable, but you can create them and delete them throughout your application like you would a regular class.

gavinray3• 1 year ago

But this is also achievable with static init methods on records and value classes, right?

  record Rational(int num, int denom) {
      Rational {
          int gcd = gcd(num, denom);
          num /= gcd;
          denom /= gcd;
      }
  }

pkulak• 1 year ago

So every time you take the hash of a string you leak 4 bytes of memory???

I assume it's static in the context of it's containing object. So, it will be collected when it's string is collected.

leksak• 1 year ago

It's similar to something like lateinit in Kotlin, except on the JVM level.

What level are you suggesting lateinit happens at if not on the JVM?

w10-1• 1 year ago

used as a static constant

Yes, but remind people it's not static in the sense of being associated with the class, nor constant for compile-time purposes.

Perhaps better to say: A stable value is lazy, set on first use, resulting in pre- and post- initialization states. The data being set once means you cannot observe a data change (i.e., appears to be immutable), but you could observe reduction in resource utilization when comparing instances with pre-set or un-set values -- less memory or time or other side-effects of value initialization.

So even if data-immutable, a class with a stable value ends up with behavior combinations of two states for each stable value. Immutable records or classes without stable values have no such behavior changes.

But, writ large, we've always had this with the JVM's hotspot optimizations.

For String, it becomes significant whether hashcode is used when calculating equals (as a fast path to negative result). If not, one would have two equal instances that will behave differently (though producing the same data), at least for one hashcode-dependent operation.

whartung• 1 year ago

I don't know, these are mostly uninformed guesses, but the distinction between Records and Value objects is that the contents lack object identity.

Which, to me, means, potentially, two things.

One, that the JVM can de-dup "anything", like, in theory, it can with Strings now. VOs that are equal are the same, rather than relying on object identity.

But, also, two, it can copy the contents of the VO to consolidate them into a single unit.

Typically, Java Objects and records are blobs of pointers. Each field pointing to something else.

With Value Objects that may not be the case. Instead of acting as a collection of pointers, a VO with VOs in it may more be like a C struct containing structs itself -- a single, continuous block of memory.

So, an Object is a collection of pointers. A Record is a collection of immutable pointers. A Value Object is (may be) a cohesive, contiguous block of memory to represent its contents.

blacklion2• 1 year ago

Records can be changed via reflection and thus doesn't participate in constant-folding in JIT phases, as this could break, for example, some serialization libraries and other nasty and dirty code. It will work in interpreter mode and eventually has heisenbugs after JIT. Not good.

`@Stable` annotation (only internal for now) and `StableValue<>` (for user code in future) says JIT that programmer guarantee (swear by his life!) that no dirty tricks are played with these values in whole codebase and JIT can constant-fold these values as soon as they are initialized.

_old_dude_3• 1 year ago

I've just tried to change a field of an instance of a record using reflection, and it's not possible, even with setAccessible(true).

_old_dude_3• 1 year ago

In Java, constants are declared as static final.

  static final Complex CONSTANT = new Complex(1, 2);

If you want a lazy initialized constant, you want a stable value

  static final StableValue<Complex> STABLE_VALUE = StableValue.of();

  Complex getLazyConstant() {
    return STABLE_VALUE.orElseGet(() -> new Complex(1, 2))
  }

If you want the fields of a constant to be constant too, Complex has to be declared has a record.

drob518• 1 year ago

A stable value, as I understand it, is either not-yet-computed or is constant. In other words, once computed it’s constant and the JIT can therefore treat it as constant.

layer85• 1 year ago

Record fields cannot be lazily initialized. The point of StableValue is lazy initialization, meaning that their value is stable if and only if they carry a non-default value (i.e. after initialization). If you don’t need lazy initialization, you can just use a regular final field. For a non-final field, without StableValue the JIT optimizer can’t tell if it is stable or not.

The implementation of a value object will be able to use StableValue internally for lazy computation and/or caching of derived values.

elric3• 1 year ago

- StableValue is about defining when and how to lazily initialy a field, with strong "exactly-once" semantics. The kinds of things we would do with safe double-checked locking before, but more convenient. Using a record doesn't address that, though you could certainly put a record in a StableValue.

Additionally, having to define a record FooHolder(Foo foo) simply to hold a Foo would be a lot more cumbersome than just saying StableValue<Foo> fooHolder = StableValue.of(); There's no need for an extra type.

- Value classes have no identity, which means they can't have synchronized methods and don't have an object monitor. While it would be possible to store a value object inside a StableValue, there are plenty of use cases for an identity object inside a StableValue, such as the Logger example inside the JEP: one could easily imagine a fictional logger having a synchronized method to preserve ordering of logs.

I wouldn't say these are all entirely orthogonal concerns, but they are different concepts with different purposes.

ajkjk• 1 year ago

It's a much-needed idea but... such an awkward way to do it. Only Java would be okay with an actual language feature using words like "orElseGet". And personally I suspect there's an inverse correlation between feature usage and how semantically awkward it is... it just feels like a mistake to even consider using an inelegant feature unless it's out of complete necessity.

It should really be something like

    public stable logger = () -> new Logger(/* .. */).

Where the JDK hides the details of making sure the value is only created once, basically like the classholder idiom but under the hood. I'm sure there are reasons why they're not doing it that way, but ... it's definitely what the language needs to be able to do.

Incidentally, I've always appreciated for python PEPs how they list all of the obvious complaints about an issue and explain methodically why each was determined not to work. The JEPs don't seem to reach quite the same candor.

Traubenfuchs2• 1 year ago

This all sounds very hard to grasp for me. Does this only work with Map.of? Would it also work with map.put?

What would be the performance improvement in average java services?

Are there specific types of applications that would benefit a lot?

Does this make string.intern() more valueable? String caches?

amiga3863• 1 year ago

Does this only work with Map.of? Would it also work with map.put?

It would be faster but not as blindingly fast. Combined with an immutable map, what it means is that the JVM can directly replace your key with its value, like the map is not even there. Because the key's hashcode won't ever change, and the map won't ever change.

Does this make string.intern() more valueable?

No, String.intern() does a different job, it's there to save you memory - if you know a string (e.g. an attribute name in an XML document) is used billions of times, and parsed out of a stream, but you know you only want one copy of it and not a billion copies). The downside is that it puts the string into PermGen, which means if you start interning normal strings, you'll run out of memory quickly.

Traubenfuchs2• 1 year ago

How does the jvm know the map is immutable?

But interned strings can also reuse their hashcode forever.

daxfohl6• 1 year ago

How would it directly replace your key with its value? What if there are bucket collisions? Do immutable maps expand until there aren't any? Moreover, what if there are hash key collisions? There needs to be some underlying mechanism to deal with these, I'd think. I don't see how replace-like-the-map-isn't-there could work. Or even how "@Stable" could be used to affect it. Would love to understand more deeply.

daxfohl6• 1 year ago

Does this make string.intern() more valuable?

Probably depends on the use case, though I'm having trouble thinking of such a use case. If you were dynamically creating a ton of different sets that had different instances of the same strings, then, maybe? But then the overhead of calling `.intern` on all of them would presumably outweigh the overhead of calling `.hash` anyway. In fact, now that `.hash` is faster, that could ostensibly make `.intern` less valuable. I guess.

daxfohl6• 1 year ago

Cool that we're still seeing perf improvements after all these years! I'm confused by some of the details in the example. Like, would we see similar 8x improvement in a simpler example like a string hashset lookup? Is there something special about MethodHandle or immutable maps here that accentuates the improvement?

Computing the hash code of the String “malloc” (which is always -1081483544)

Makes sense. Very cool.

Probing the immutable Map (i.e., compute the internal array index which is always the same for the malloc hashcode)

How would this work? "Compute" seems like something that would be unaffected by the new attribute. Unless it's stably memoizing, but then I don't quite see what it would be memoizing here: it's already a hash map.

Retrieving the associated MethodHandle (which always resides on said computed index)

Has this changed? Returning the value in a hash map once you've identified the index has always been zero overhead, no?

Resolving the actual native call (which is always the native malloc() call)

Was this previously "lazyinit" also? If so, makes sense, though would be nice if this was explained in the article.

twic• 1 year ago

How would this work? "Compute" seems like something that would be unaffected by the new attribute.

The index is computed from the hashcode and the size of the array. Now that the hash code can be treated as a constant, and the size of the array is already a constant, the index can be worked out at compile time. The JVM can basically inline all the methods involved in creating and probing the map, and eliminate it entirely.

daxfohl6• 1 year ago

The bucket is computed from the hash code and the size of the array, but that's not necessarily the index. If there are no bucket collisions, then index==bucket and this works out. But if there are bucket collisions then the index will be different from the bucket. So you still need some computation IIUC. And there's no way to memoize that result, since memoization would require a hashmap that has the exact same characteristics as the original hashmap.

I guess a @Stable attribute on the array underlying the map would allow for the elimination of one redirection: in a mutable map the underlying array can get resized so its pointer isn't stable. With an annotated immutable map it could be (though IDK whether that'd work with GC defrag etc). But that seems like relatively small potatoes? I don't see a way to "pretend the map isn't even there".

esafak• 1 year ago

Will Kotlin and Scala benefit?

daxfohl6• 1 year ago

They should. It's a JVM feature on core JDK classes.

taeric2• 1 year ago

This implies they have their own String implementations? I wouldn't be shocked to find out that is the case, but it would surprise me.

jillesvangurp3• 1 year ago

If you use Kotlin/Scala on the JVM, yes. You can also use Kotlin on native (ios, win32, linux, etc), wasm, or on top of javascript. Scala has at least a js compiler and maybe some other ones as well. You won't see any improvements on these other platforms.

_tom_• 1 year ago

Off-the-cuff thought:

Could you solve the empty string hashes to zero problem by just adding one when computing hash codes?

layer85• 1 year ago

You could, but that would break backwards compatibility.

Karellen2• 1 year ago

But then strings with the hash code HASH_MAX would wrap to 0 instead.

Bootvis• 1 year ago

Is @Stable a generalization of the final keyword or is there more to it that I'm missing?

selkin• 1 year ago

Final is an access modifier. It controls who can change a value (in this case, no one). And like other access modifiers, you can use reflection to remove or change those modifier. It is a security barrier the developer asks the system to uphold.

The Stable annotation is an optimization mechanism: a promise the developer makes to the compiler. It is on the developer to uphold.

vault• 1 year ago

why is a website about programming not writing code blocks in ASCII? I'm referring to quotes and other undesirable symbols

VWWHFSfQ• 1 year ago

Java supports non-ASCII source code since forever. A lot of languages do.

dullcrisp• 1 year ago

But it doesn't support using fancy unicode quotes as string delimiters.

int_19h4• 1 year ago

I'm rather surprised that string hashes aren't randomized in JVM - or at least that's what the article implies. These days, it's a fairly common mitigation for DDoS attacks based on hash collisions. Does Java not do it to preserve backwards compatibility because too much existing code relies on a specific hashing algorithm?

sedatk• 1 year ago

Java uses a tree instead of a linked list for collided items, so the search performance degrades more gracefully (e.g. O(N) vs O(logN)).

77• 1 year ago

To be more precise, Java initially uses a linked list for nodes within a bin. If the number of items inside the bin crosses TREEIFY_THRESHOLD (which is 8), then that specific bin is converted into a RB tree.

This is detailed in implementation notes comment here: https://github.com/openjdk/jdk/blob/56468c42bef8524e53a929dc...

jillesvangurp3• 1 year ago

Depends on which Map / Set implementation you use. There are multiple, each with different and interesting properties.

yxhuvud• 1 year ago

Nothing stops the jvm from caching hashes even if the hashes are unique per process invocation.

PaulHoule• 1 year ago

They aren’t and it’s quite unfortunate that the empty string hashes to 0 so it will have to recompute every time although presumably it is quick to compute the hash of the empty string.

ivan_gammel16• 1 year ago

Java hashCode contract is to optimize calculation of hash for performance not for collision search resistance. Its sole purpose is to use it in collections. It must not be used in situations where you need cryptographic properties.

tialaramex6• 1 year ago

So, the problem here is that you're thinking of "cryptographic properties" as only the "cryptographic hashes" such as those built with the Merkle–Damgård construction (SHA-512/256 being the one you might reasonably pick today)

But, it's actually desirable to have some cryptographic properties in a faster one way function for making hash tables. Read about SipHash to see why.

Because Java didn't (and as others have discussed, now can't) choose otherwise the hash table structures provided must resist sabotage via collision, which isn't necessary if your attacker can't collide the hash you use.

GolDDranks• 1 year ago

The thing is, even it being used just in collections can lead to DOS, if the attacker can control the string contents, and selectively choose strings that your hash table stops being a hash table and turns into a linked list.

andrewaylett• 1 year ago

Even in collections, an unstable hash is desirable -- to avoid denial of service attacks caused by attacker-controlled hash collisions.

For example, Python back in 2012: https://nvd.nist.gov/vuln/detail/cve-2012-1150.

ncruces• 1 year ago

The hashing algorithm is specified in the docs:

   s[0]*31^(n-1) + s[1]*31^(n-2) + ... + s[n-1]

Per Hyrum's law, there's no changing it now.

https://docs.oracle.com/javase/8/docs/api/java/lang/String.h...

jillesvangurp3• 1 year ago

No, it would break the semantics of equals and hash which dictate that two objects that are equal should also have the same hash code. So hash codes for objects must be deterministic. Which in turn is an important property for sets, hash tables, etc. It's unlikely to be ever changed because it would break an enormous amount of stuff.

For things that need to be secure, there are dedicated libraries, standard APIs, etc. that you probably should be using. For everything else, this is pretty much a non issue that just isn't worth ripping up this contract for. It's not much of an issue in practice and easily mitigated by just picking things that are intended for whatever it is you are trying to do.

xmcqdpt2• 1 year ago

Languages that choose to fix this problem at the hash code level have hash codes that are still deterministic within a given program execution. They are not deterministic between executions, cf

https://docs.python.org/3/reference/datamodel.html#object.__...

stevoski2• 1 year ago

I find the entire history of improvements to Java’s String class enjoyable to read about.

Over the years, the implementation of Java’s String class has been improved again and again, offering performance improvements and memory usage reduction. And us Java developers get these improvements with no work required other than updating the JRE we use.

All the low-hanging fruit was taken years ago, of course. These days, I’m sure most Java apps would barely get any noticeable improvement from further String improvements, such as the one in the article we’re discussing.

niuzeta5• 1 year ago

I love hearing more about this, especially the historical context, but don't have a good java writeups/articles on this. Would you mind sharing some suggestions/pointers? I'd very much appreciate it.

stevoski2• 1 year ago

A good starting point is Joshua Bloch’s Effective Java. He shares some stories there from Java’s early days, and - at least in passing - mentions some aspects of the String class’s history.

DaiPlusPlus• 1 year ago

Would you mind sharing some suggestions/pointers?

I would, but unfortunately I got a NullPointerException.

I suggest you try Rust instead; its borrow checker will ensure you can't share pointers in an unsafe manner.

wging• 1 year ago

String compression was one. tl;dr: the JVM supports Unicode for strings, but uses 1-byte chars for strings where possible (previously it was UTF-16), even though it's not actually doing UTF-8.

Depending on what sort of document you're looking for, you might like either the JEP: https://openjdk.org/jeps/254

or Shipilev's slides (pdf warning): https://shipilev.net/talks/jfokus-Feb2016-lord-of-the-string...

Shipilev's website (https://shipilev.net/#lord-of-the-strings), and links from the JEP above to other JEPS, are both good places to find further reading.

(I think I saw a feature article about the implementation of the string compression feature, but I'm not sure who wrote it or where it was, or if I'm thinking about something else. Actually I think it might've been https://shipilev.net/blog/2015/black-magic-method-dispatch/, despite the title.)

cempaka4• 1 year ago

This is a good video that goes over a ton of the optimizations, especially around concatenation: https://youtu.be/z3yu1kjtcok?si=mOdZ5sh5rp8LNyap

neuroelectron4• 1 year ago

When I started my career in software development, SDE, and soon advanced to SRE, I hated Java. The extreme OOP paradigm made enterprise class situations impossible to understand. But after a few short years, I began to appreciate it as a real, battle hardened ecology. Now, I consider it much better than modern trends such as Rust and Python.

These kinds of niche optimizations are still significant. The OOP model allows them to be implemented with much less fanfare. This is in the context of billion-dollar platforms. With some basic performance testing and API replays, we're saving thousands of dollars a day. Nobody gets a pat on the back. Maybe some pizza on Friday.

jimmaswell• 1 year ago

I've seen Java described as made for companies to be able to rotate out mediocre programmers as efficiently as possible without letting them mess things up easily, and it makes a lot of sense from that perspective. Barebones semantics to the point of being Spartan (can't even define your own operator overloads), easy to take another class and copy it with small modifications but not mess it up for anyone else (inheritance)..

Then there's C# which most anyone who's enthusiastic about software dev will find far nicer to work with, but it's probably harder for bargain basement offshore sweatshops to bang their head against.

threeseed• 1 year ago

You can use the JVM without needing to use OOP e.g. Scala, Clojure, Python, Javascript, Ruby etc.

Then you can get to benefit from Java's unparalleled ecosystem of enterprise hardened libraries, monitoring etc.

SkiFire133• 1 year ago

Did you actually started to appreciate the same OOP that made class situations impossible to understand or did you gradually switched to a simplier OOP, made up of mostly interfaces and classes that implement them (as opposed to extending other classes)?

In my experience OOP is actually pretty pleasant to work with if you avoid extending classes as much as possible.

These kinds of niche optimizations are still significant. The OOP model allows them to be implemented with much less fanfare.

If you're referring to the optimization in the article posted then I would argue an OOP model is not needed for it, just having encapsulation is enough.

davnicwil• 1 year ago

the mind blowing moment for me with Java came about 5 years into using it, when I encountered the idea - via some smart colleagues - that none of the extra 'stuff' is intrinsic to the language but rather is self-imposed.

Turns out you can write java without the stuff. No getters and setters, no interfaces or dependency injection, no separate application server (just embed one in your jar). No inheritence. Indeed no OOP (just data classes and static methods).

Just simple, c-like code with the amazing ecosystem of libraries and the incredibly fast marvel that is the JVM and (though this is less of a deal now with LLM autocomplete) a simple built in type system that makes the code practically write itself with autocomplete.

It's truly an awesome dev experience if you just have the power / culture to ignore the forces pressuring you to use the 'stuff'.

bradhe• 1 year ago

no work required other than updating the JRE we use

Have you tried updating production usage of a JRE before??

microflash• 1 year ago

I have done it multiple times for different versions of Java with nominal effort. Of course, difficulty may vary depending on a project.

With projects like OpenRewrite [1] and good LLMs, things are a lot easier these days.

[1] https://docs.openrewrite.org

znpy• 1 year ago

Yes. I moved a few repository from java 8 up to Java 21.

Java 8 -> 9 is the largest source of annoyances, past that it's essentially painless.

You just change a line (the version of the JRE) and you get a faster JVM with better GC.

And with ZGC nowadays garbage collection is essentially a solved problem.

I worked on a piece of software serving almost 5 million requests per second on a single (albeit fairly large) box off a single JVM and I was still seeing GC pauses below the single millisecond (~800 usec p99 stop the world pauses) despite the very high allocation rate (~60gb/sec).

The JVM is a marvel of software engineering.

paulddraper• 1 year ago

The flip side is that for years Java developers have been dealing with suboptimal strings with nothing to do about it.

spoonsort• 1 year ago

Having a @Stable annotation which the standard library can use to signal things to the compiler like that is very smart. I wonder how many standard libraries are missing out on optimizations like this.

delusional2• 1 year ago

The example is entirely unconvincing. Why would you store those calls in a map and not just a variable?

Even if the map is crucial for some reason, why not have the map take a simple value (like a unint64) and require the caller to convert their string into a slot before looking up the function pointer. That way the cost to exchange the string becomes obvious to the reader of the code.

I struggle to find a use case where this would optimize good code. I can think of plenty of bad code usecases, but are we really optimizing for bad code?

koolba• 1 year ago

I struggle to find a use case where this would optimize good code. I can think of plenty of bad code usecases, but are we really optimizing for bad code?

The most common such usage in modern web programming is storing and retrieving a map of HTTP headers, parsed query parameters, or deserialized POST bodies. Every single web app, which arguably is most apps, would take advantage of this.

delusional2• 1 year ago

storing and retrieving a map of HTTP headers.

I dont have the profiling data for this, so this is pure theoretical speculation. At the time you're shoving http headers, which is dynamic data that will have to be read at runtime, into a heap allocated datastructures inside the request handling. It kinda feel like doing a little xor on your characters is a trivial computation.

I don't envision this making any meaningful difference to those HTTP handlers, because they were written without regard for perfomance in the first place.

f33d5173• 1 year ago

but are we really optimizing for bad code?

Yes

gf0007• 1 year ago

I think it's a pretty myopic view - this exact case might not appear as is, but might readily appear in completely sane and common code after inlining a few things and other optimisations.

joejev• 1 year ago

Isn't the entire point of an optimizer to convert "bad code" into "good code"?

Your proposed solution is to have the user manually implement a hash table, but if you have a good optimizer, users can focus on writing clear code without bugs or logic errors and let the machine turn that into efficient code.

ashvardanian2• 1 year ago

Has anyone done/shared a recent benchmark comparing JNI call latency across Java runtimes? I’m exploring the idea of bringing my strings library to the JVM ecosystem, but in the past, JNI overhead has made this impractical.

throwaway2037• 1 year ago

Can you share a link to your "strings library"? I am curious about what it can do that a Java String cannot.

ashvardanian2• 1 year ago

At this point, it doesn’t provide much novel functionality, but it should be faster than the standard libraries of most (or maybe all) programming languages.

https://github.com/ashvardanian/StringZilla

cempaka4• 1 year ago

Java has replaced JNI with the Project Panama FFM, which depending on your use case might perform quite a bit better than JNI used to. The Vector API is stuck in incubator and still a bit rough around the edges though, so SIMD might be a bit trickier.

kazinator2• 1 year ago

I experimented with hash codes in strings in TXR Lisp, but took it out after several days:

  commit a136f37015cc2513878f75afcf8ba49fa61a88e5
  Author: Kaz Kylheku <kaz@kylheku.com>
  Date:   Sat Oct 8 20:54:05 2022 -0700

    strings: revert caching of hash value.

    Research indicates that this is something useful in
    languages that abuse strings for implementing symbols.
    We have interned symbols.

    * lib.h (struct string): Remove hash member.

    * lib.c (string_own, string, string_utf8, mkustring,
    string_extend, replace_str, chr_str_set): Remove
    all initializations and updates of the removed
    hash member.

    * hash.c (equal_hash): Do not cache string hash value.

> This improvement will benefit any immutable Map<String, V> with Strings as keys and where values (of arbitrary type V) are looked up via constant Strings.

Wait, what? But, that's inherently constant foldable without reasoning about string hash codes; we don't need them at all.

We examine the expression [h "a"]: lookup the key "a" in hash table h, where h is a hash literal object, that we write as #H(() ("a" "b)). It contains the key "a", mapping it to "b":

  1> (compile-toplevel '[#H(() ("a" "b")) "a"])
  #<sys:vm-desc: 8eaa130>

What's the code look like?

  2> (disassemble *1)
  data:
      0: "b"
  syms:
  code:
      0: 10000400 end d0
  instruction count:
      1
  #<sys:vm-desc: 8eaa130>

One instruction: just return "b" from the static data register d0. The hash table is completely gone.

The keys don't even have to be strings; that's a red herring.

josefx• 1 year ago

Wait, what? But, that's inherently constant foldable without reasoning about string hash codes; we don't need them at all.

Their goal wasn't to improve key lookups in hash tables, that is more or less just an example.It was to improve optimization of variables with lazy initialisation overall and the hash of String uses lazy initialisation.

jbverschoor• 1 year ago

So strings don’t get hash code at compile time?

At first I thought the article was describing something similar to Ruby’s symbols

chii2• 1 year ago

So strings don’t get hash code at compile time?

only strings that are known at compile time could possibly be compile-time hashed?

But the article is talking about strings in a running program. The performance improvements can apply to strings that are constants, but is created at run time.

layer85• 1 year ago

It’s a bit unfortunate that the user code equivalent (JEP 502) comes at the cost of an extra object per “stable” field. Lazy initialization is often motivated by avoiding creating an object up-front, but with this new pattern you’ll have to create one anyway.

elric3• 1 year ago

Well, no. The JVM makes the wrapper object disappear. One of the design drivers for StableValue was performance.

I mean the developer has to create the StableValue field, but its access is optimized away.

layer85• 1 year ago

After JIT compilation, yes (presumably), but for interpreted byte code I assume that a regular object is still allocated.

rf15• 1 year ago

Strange to make an entire piece on String specifically - it seems like this is just based on the work done for the @Stable annotation and the many effects it can have on classes during runtime.

vanderZwan• 1 year ago

I can understand the choice as an example for @Stable: it highlights that this addition even comes with "free" upgrades for (some) existing code out there, and it immediately shows a concrete use case. But yes, they could have gone further into generalizing from there.

If you mean the headline, Strings are a universal data type across programming, so claims of improving their performance gets more clicks than "this annotation that you have never heard about before makes some specific code faster", especially when it comes to getting the attention of non-Java programmers.

theanonymousone• 1 year ago

It seems string typing is not as bad an anti-pattern as it used to be :)

It will be a very impactful work; I'm excited to see. Probably even a 1% improvement in String::hashCode will have an impact on global carbon footprint or so.

Karellen2• 1 year ago

Um, malloc() is not a system call?

xxs2• 1 year ago

no, mmap is a system call; the memory allocators tend not to use syscalls (often at all) as objection instantiation is very common; also it has to be concurrent and what not.

koliber• 1 year ago

Is this a correct summary: we memoized the hashCode() function?

matsemann• 1 year ago

Not only that. By making it guaranteed constant after first invocation, they can fold the constant and basically avoid the map lookup in runtime and instead do it compile-time. In practice avoiding the map entirely.

nimrody2• 1 year ago

No. String.hashCode() was already memoized. So after the first call to hashCode(), future calls just retrieved the data from the hash field of the string object.

This optimization is about avoiding even calling the method because the jvm knows that the value returned will be the same.

smcin• 1 year ago

Notwithstanding HN's guidline to preserve original title and not editorialize in any way, this would be less misleading if it had been titled 'Java Strings Just Got Faster in JDK 25'

Strings Just Got Faster

Related Stories