Reg: Proposal: Generalized Number.parseNumber(String) Method for Java

[Joseph D. Darcy]

Another reaction: in practice there is not much numeric about 
java.lang.Number. Operationally the type means "convertible to a 
primitive type" and there is nothing else extending Number lets you do 
with a class. Additionally in retrospect, Number would have been better 
as an interface rather than an abstract class.

When looking to augment numeric capabilities of the platform, directly 
involving java.lang.Number is not the first or second place I'd look.

Cheers,

-Joe

On 3/28/2025 11:38 AM, Kevin Bourrillion wrote:
> I appreciate you raising this; I think it is relevant to some internal 
> discussions we’re having about the future of Number. Here’s some 
> personal reactions from a team member (not a team consensus opinion):
>
> Unfortunately, we’ve found many times that `Number` is a pretty 
> deficient return type that users can’t do much of anything useful with 
> (except apply arbitrarily lossy conversions to, or test with 
> `instanceof`).
>
> And I’m skeptical that most users ever want to end up in a situation 
> where they have a `Number[]` or `Collection<Number>` that could be 
> heterogeneous. This state feels like a temporary one you want to get 
> out of.
>
> It’s not clear that `Number` can really be rehabilitated much, either. 
> It is sort of a “deficient type” by its nature.
>
> When you do need this sort of flexible parsing, I think the status quo 
> is: you can always parse to BigDecimal first, then ask questions about 
> what more “minimal” type the value you got might fit into. Obviously 
> this does not have optimal performance, but I am not sure there’s a 
> good spot-fix here that doesn’t just raise more issues than it tried 
> to resolve.
>
> We will keep thinking about this, though. Valhalla means we will 
> inevitably have more "numeric types" than ever, and this is prompting 
> us to think about what these types should have in common. (It’s a big 
> topic and I apologize for not even attempting to outline it all for 
> you here yet.)
>
>
>
>> On Mar 28, 2025, at 10:29 AM, Sathish Kumar Thiyagarajan 
>> <sathishkumar.thiyagarajan at gmail.com> wrote:
>>
>> Dear Core-Libs Dev Team,
>>
>> *Note:* I have now subscribed to the mailing list and am resending 
>> this message as advised.
>>
>> I would like to propose an improvement to the JDK: a *generalized 
>> |Number.parseNumber(String)| method*.
>>
>>
>>       Motivation
>>
>> Java provides multiple number types (|byte|, |short|, |int|, |long|, 
>> etc.), and developers typically choose them based on memory 
>> considerations. Currently, Java offers |String| to |Number| 
>> conversions using concrete classes:
>>
>>  *
>>
>>     |Long.parseLong(String)|
>>
>>  *
>>
>>     |Integer.parseInt(String)|
>>
>>  *
>>
>>     |Short.parseShort(String)|, etc.
>>
>> While these are useful, Java lacks a *generalized method* that 
>> returns the most memory-efficient |Number| representation based on 
>> the input, like:
>>
>> |Number.parseNumber(String numberAsText); |
>>
>>
>>       Use Case: JSON Serialization
>>
>> This would be particularly useful in cases like *JSON serialization 
>> in REST APIs (Using Jackson <https://github.com/FasterXML/jackson>)*, 
>> where number types are often altered during 
>> serialization/deserialization. Consider the following test case:
>>
>> |@Test void testNumberMemoryUsage() throws JsonProcessingException { 
>> ObjectMapper mapper = new ObjectMapper(); Map<String, Object> 
>> numbersObject = Map.of("aShort", (short) 1234, "aFloat", (float) 
>> 1.33); final String jsonText = 
>> mapper.writeValueAsString(numbersObject); Map<String, Object> 
>> parsedJsonObject = mapper.readValue(jsonText, new TypeReference<>() 
>> {}); // Expected: Short.class | Actual: Integer.class 
>> assertEquals(Short.class, parsedJsonObject.get("aShort").getClass()); 
>> // Expected: Float.class | Actual: Double.class 
>> assertEquals(Float.class, parsedJsonObject.get("aFloat").getClass()); } |
>>
>>
>>       Reference Implementation
>>
>> Here’s a rough implementation to illustrate the idea:
>>
>> |private static Number parseNumber(final String numberStr) { try { if 
>> (numberStr.contains(".")) { double doubleValue = 
>> Double.parseDouble(numberStr); return (doubleValue >= 
>> -Float.MAX_VALUE && doubleValue <= Float.MAX_VALUE) ? (float) 
>> doubleValue : doubleValue; } else { long longValue = 
>> Long.parseLong(numberStr); if (longValue >= Byte.MIN_VALUE && 
>> longValue <= Byte.MAX_VALUE) { return (byte) longValue; } else if 
>> (longValue >= Short.MIN_VALUE && longValue <= Short.MAX_VALUE) { 
>> return (short) longValue; } else if (longValue >= Integer.MIN_VALUE 
>> && longValue <= Integer.MAX_VALUE) { return (int) longValue; } else { 
>> return longValue; } } } catch (NumberFormatException e) { return 
>> parseBigNumber(numberStr); } } private static Number 
>> parseBigNumber(final String numberStr) { try { return new 
>> BigInteger(numberStr); // Try BigInteger first } catch 
>> (NumberFormatException e) { // Only create BigDecimal if BigInteger 
>> fails BigDecimal bd = new BigDecimal(numberStr); try { // Convert to 
>> BigInteger if there's no fraction return bd.toBigIntegerExact(); } 
>> catch (ArithmeticException ex) { return bd; // If it's a decimal, 
>> return BigDecimal } } } |
>>
>> Would love to hear your thoughts on this proposal. Appreciate your 
>> feedback and guidance!
>>
>> Thanks & Regards,
>> Sathish Kumar Thiyagarajan
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/core-libs-dev/attachments/20250328/e26b7d57/attachment.htm>