From Dev to Deploy Without Leaving Your Editor

How Softanza models the whole system journey as one program

Underneath the tool sprawl is a subtler tax: a set of invisible frames the programmer must carry in their head and re-supply on every line —

• Which system does this code run on? (my laptop now, the device later)
• What is that system allowed to do? (an MCU has no processes; a phone is sandboxed)
• Will this even run there? (I have no GPIO on my laptop)
• How do I get it there? (build, cross-compile, flash, provision)

Softanza's System Foundation makes each of those frames a named, checkable thing — and models the whole dev-to-deploy problem as a single object graph in one vocabulary. The result is that the journey from your editor to a running device can be one continuous program, not seven disconnected tools. This narration walks that program end to end. Every code block below is real, and every output block is its actual output.

One developer, one product

Meet a developer on a Windows laptop building a smart greenhouse: a Linux backend, an Android companion app, and an ESP32 sensor node that reads soil moisture and drives a pump. Here is their whole dev-to-deploy session.

1. Who am I developing on?

No guessing, no ambient "current machine." The development system is an object you instantiate, read live from the engine:

o1 = new stzDevSystem()

? o1.OSName() + " / " + o1.Architecture() + " / " + o1.BitSize() + "-bit"
#--> windows / x64 / 64-bit

? @@( o1.Capabilities() )
#--> [ "filesystem", "process", "network", "environment", "dynamic_load", "threads", "clock" ]

? o1.Lacks(:gpio)
#--> 1

That last line matters more than it looks: the machine's capabilities are first-class data. The object knows the laptop has no GPIO — and it will use that fact in a moment to help, not to block. (CurrentSystem() — a new stzCurrentSystem() — is its runtime-scope twin: on a deployed device it reads that device.)

2. Model the whole product once

The deployment architecture is declared data, in four lines, in one place — not scattered across a Dockerfile, a compose file, and a CI pipeline:

oProduct = new stzPlatformProfile("smart-greenhouse")
oProduct.DevelopedOn(:Windows)
oProduct.AddServer(:backend, :LinuxServer)
oProduct.AddSuperApp(:phone, :Android)
oProduct.AddApp(:sensor, :ESP32)

Platform: smart-greenhouse
  developed on: windows
  server 'backend' -> linux
  superapp 'phone' -> android
  app 'sensor' -> espidf

Each part has a kind (server / superapp / app) and a target system. The profile only describes the solution; nothing is built or deployed yet.

3. Write feature code in a named scope

This is where the tax disappears. Feature code for a part is authored against that part's target, and checked on the laptop, as it is typed:

? oProduct.ReadPinIn(:sensor, 34)      # read the moisture pin
? oProduct.WritePinIn(:sensor, 26, 1)  # drive the pump
oProduct.SpawnIn(:sensor, "worker")    # ... and try to spawn a process

ReadPin(34)    -> rehearsed   (ESP32 allows gpio; rehearsed -- my laptop has none)
WritePin(26,1) -> rehearsed
Spawn('worker') -> REFUSED here, on my laptop, at write time
                   (an MCU has no processes -- I learn it NOW, not on the bench)
backend Spawn('cron') -> native   (Linux + my laptop both have processes)

Read those four results as a programmer, not a spec:

• Spawn on the sensor was refused — on the laptop, at write time. In a conventional flow you learn that an MCU has no processes after cross-compiling and flashing, staring at a dead board, or from a field crash. Here the whole class of "the target won't allow that" errors moved from production to your editor. This is shift-left applied to the deployment target itself.

• ReadPin came back rehearsed, not refused. The ESP32 does have GPIO; the laptop doesn't. Rather than fail, the scope rehearses the operation — so you can write and reason about firmware for hardware you do not physically have. The absent world is made present.

• The same Spawn is native on the backend. The scope, not the operation, decides. One verb, three verdicts, each correct for its target.

4. A risky system change? Rehearse it first

System operations that touch reality — reorganising files, changing the environment, spawning children — go through a twin: an in-memory rehearsal that holds no reference to reality. You try things, read a plain-language plan, and reality changes only when you commit it:

oVfs = new stzVirtualFileSystem()
oVfs.CreateFolder("config")
oVfs.CreateFile("config/greenhouse.ini", "pump_threshold=40")
oVfs.MoveFile("old_settings.txt", "config/legacy.ini")
oPlan = oVfs.GenerateUpdatePlan()
? oPlan.Narration()

Update plan (3 of 3 operations to commit):
  * 1. create folder 'config'
  * 2. create file 'config/greenhouse.ini' (17 bytes)
  * 3. move 'old_settings.txt' -> 'config/legacy.ini'

reality is untouched until I choose to commit this plan.

System programming stops being "one wrong command and it's gone."

5. Build, then Deploy

A real stzPlatform owns the profile and drives its lifecycle. We do not deploy a profile — we deploy a platform and its parts, and building precedes deploying as two honest, ordered steps:

oPlatform = StzPlatformQ("smart-greenhouse")
oPlatform.SetProfile(oProduct)
oPlatform.Build()
oPlatform.Deploy()

Build()  -> built 3 parts
Deploy() -> lowered + delivered each part to its target

6. The firmware — written on a laptop with no GPIO

At deploy, the lowering bridge turns each part's rehearsed operations into a real target artifact. The GPIO code the developer wrote on a Windows laptop becomes actual firmware source:

? oPlatform.ArtifactFor(:sensor)

// firmware generated by the Softanza lowering bridge
// target: espidf
void setup() {}
void loop() {
  digitalRead(34);
  digitalWrite(26, 1);
}

ReadPin(34) lowered to digitalRead(34); WritePin(26, 1) to digitalWrite(26, 1). The developer validated firmware logic without ever holding the board. The board is now the last inch, not the prerequisite for any progress.

7. The backend, cross-compiled for real — from the same laptop

The lowering bridge writes the source; stzBuilder turns it into a real binary for the target. Softanza's engine is built with Zig, and Zig is a complete multi-platform cross-compiler — so the Linux backend builds from the Windows laptop with no separate toolchain. And the build target is not re-specified: it is the part's deployment system, the same object that governed what the code may do:

oBuild = new stzBuilder("backend")
oBuild.AddSourceQ("backend.c")
oBuild.SetTargetQ( oProduct.App(:backend).DeploymentSystem() )   # a system IS a build target
oBuild.ReleaseFast()
oBuild.Build()

? oBuild.Target()
#--> x86_64-linux-gnu

? oBuild.Succeeded()
#--> 1

The output is a genuine Linux ELF executable, produced on Windows — the same laptop with no GPIO, no Linux, and now no Linux toolchain either. The deployment system that decided what the code may do is the very object that decides what it compiles to. One profile carries the whole journey, from "which machine?" to a shippable binary.

8. Hand it to an AI copilot

The very same capability model that decided "can the ESP32 spawn?" decides "can this actor deploy?" Deployment is an effectful crossing, so an actor whose effect-capability set is empty — an LLM — may propose it but cannot perform it:

oGuarded.DeployAs(LLMActor("copilot"))   # raises
oGuarded.DeployAs(HumanActor("me"))      # proceeds

the copilot may PROPOSE the deploy -- but CANNOT perform it (no effect capability)
I approve; the human deploys. Expression is free; admission is governed.

You wrote no bespoke permission plumbing. Safety was a property of the model you were already using.

What changed, frame by frame

The deepest simplification: the seam collapses

"It worked in dev, it broke in prod" is a law of nature in conventional stacks for one reason: dev and deploy are two different worlds joined by a fragile hand-off — different tools, different config, different invisible assumptions, discovered to disagree only at the boundary.

Softanza makes deploy a first-class part of the dev-time model. You write, check, build, and deploy against the same profiles, with the target's real constraints enforced on your machine before the hand-off even exists. The cliff between the two worlds becomes a named scope boundary you can see and check. That is the whole trick, and it is why the session above reads like one continuous thought rather than a relay race between seven tools.

There is a unifying reason it holds together: one vocabulary runs through the entire pipeline. The capability lattice — effectful / sensing / compute / inference — is the same word used for a system's capabilities, a deployment scope's verdicts, an actor's authority, and the deploy governance gate. The developer above learned four nouns — a capability, a scope, an actor, a plan — once, and they carried from "what machine am I on?" to "here is the firmware" to "the AI can't deploy it." The mental model never fragmented.

Honest about the edges

A good tool is precise about what is real and what is modelled:

• The flash / upload of the artifact to the physical device is genuinely external — it needs the hardware. The model generates the firmware; the last inch is real, and stays honest about it.
• The cross-compile is real: stzBuilder drives Zig, a genuine multi-platform C/C++/Zig compiler — the Linux (and wasm) binaries above are verified by their ELF/wasm magic bytes, built from Windows. What stays modelled is the ESP32 firmware source (digitalRead is a faithful Arduino stand-in); a real MCU build plugs an ESP-IDF / arduino-cli toolchain behind the same builder interface. The seam is unified; that one target still awaits its real tool.
• This does not replace every mature deploy tool. It unifies the seams those tools leave exposed — which is where the cognitive tax actually lives.

The claim that survives running it: Softanza does not make deployment more powerful so much as it makes the dev-to-deploy problem small enough to hold in one head — because it is one model, one vocabulary, checked early, on your own machine.

See also

• ../design/SOFTANZA_SYSTEM_FOUNDATION.md — the full design: engine-true facts, the three scopes, the twin, the governance crossing, and the lowering bridge.
• ../design/SCOPE_ORIENTED_PROGRAMMING.md — the paradigm underneath: name the operative scope at the call site and let the library reason with it (regex was the first governed field; system programming is the second).
• stz-lowlevel-system-programming.md — the low-level vision (safe memory, named cross-platform operations) this foundation sits on.

Citer

Mansour Ayouni. « From Dev to Deploy Without Leaving Your Editor ». Softanza narrations, 2026. https://mayouni.github.io/stzsite/fr/narrations/stz-system-dev-to-deploy-narration.html

Exécuté au nom de Haro le 2026-10-08 20:42 dans la bibliothèque au commit 4a184e6f0, bloc après bloc dans un seul processus : 1 bloc(s) sur 8 ont tenu leur promesse ; les autres sont des illustrations. Le code des articles est du code Haro. Là où un article a été écrit avant que la langue prenne son nom actuel, le nom a été changé et rien d'autre. Genre, domaine et série lus dans le nom du fichier, année tirée du premier commit : dérivés, jusqu'à ce que l'article porte son propre en-tête. Le fichier sur GitHub(s'ouvre dans un nouvel onglet)