The architecture began in 2005, in an open-source project rather than a client system.
jWebDBAdmin was published on
SourceForge with a single design goal: stay lightweight. One Portal
servlet resolved a command name to a bean, populated that bean from the request, ran it, and handed the result to one
template. Spring 1.0 was about a year old at the time and had not yet become the default it later was — this was a choice
made before the standard settled, not one made against it.
When Gaharu needed a new support system in 2009, that foundation had already proven itself in the open. Kaliber was built
on it rather than on something new. Every re-evaluation since has been a genuine one: Spring, Maven, and Gradle have each
been revisited several times, most recently in 2024, and the conclusion has held.
The trade-off is close to even, and that is precisely the reason it holds. When something breaks in code written here, that
code gets read. When something breaks inside a framework, the fix requires learning that framework's internals to solve a
problem that was never yours — and the timeline belongs to somebody else. The same applies to build tools: Maven would mean
maintaining Maven, rather than the library actually needed.
None of it was free. What a framework would have supplied had to be written instead, and that became GeneralLib —
grown incrementally, driven by need rather than designed up front. Its database layer is the clearest example: a connection
pool that validates and recycles connections, and a wrapper implementing
java.sql.Connection so it drops in anywhere, tracks every
statement it creates, and closes all of them when released. Two classes remove the entire leaked-cursor class of bug, with
no ORM anywhere. The SMS layer started from a GPL PDU library, extended for UDH ports, alternative character encodings, and
custom acknowledgement; the SMTP and TCP servers were written here.
The payoff is visible in the lineage: jWebDBAdmin (2005) → Kaliber (2009) → Saturn → yakobusOne → AppKTG → aether.ing (2025).
The presentation layer changed completely along the way, from HTTP request/response to a single-page application. The core
did not — twenty years on, aether.ing still resolves a request through a
Portal, a
Command, and a
BeanObject, the same three pieces the 2005 source defines.
Kaliber has never been fundamentally rebuilt; modules were added throughout, and its most notable change in seventeen years
was Indonesia's VAT moving to 11%, then to 12% applied as 11/12 of the base.
The cost, honestly. The architecture is uncommon, so a developer joining it has to learn it — historically the real
price of this decision. The same decision holds today all the same, and that objection has weakened: AI coding assistants
absorb an unfamiliar codebase's idiom quickly, so an uncommon framework costs a newcomer far less than it used to. What
still matters is the architecture underneath — the part that decides whether a system can flex and scale.
SMPP
SNMP
SIP
WebSocket
OpenID
serial tty