FiveTech Support Forums

Absolutely, It will take a lot of thinking, piece by piece, lots of copy and paste, lots of thinking on how to make it stick, but you have to simplify, use what is there, know both worlds, what is possible, and what is not.... and there are battles you take, and some you don't.... it will be bugs by the millions... some things you get Grok to do, some Zencoder, some MyNinja.AI, Claude.ai, Goodai, so you learn what to give to whom....

I see my last post was 5 years ago... so its been a while, i was creating the functions, and lots of flow charts, thinking how to tie it together, making classes, reading documentations, creating security, building the platform, attaching frameworks for debugging, windows api translation tries, lots of codeigniter and laravel, books, so yes it took a lot effort and time, before i could do the first test

Hello friends,

Workflow “Migration with Preprocessor instead of OOP”

Capture the screen layout

Existing dialogs or masks are provided as screenshots to the AI.

This way, the AI recognizes the visual structure: fields, buttons, tables.

Provide the legacy code

At the same time, the old PRG source (e.g. SAY, GET, BUTTON, BROWSE) is added.

From this, the AI extracts the logic: field names, validations, conditions, actions.

Output: Web code without OOP

The AI generates HTML, CSS and JavaScript that run directly in the browser.

No attempt to rebuild desktop-style window classes – instead, native web elements (forms, tables, buttons).

Server logic is delivered as lean PHP functions, not a forest of classes.

Integrate preprocessor/patcher

During generation, markers like -> HEADER, -> NAV, $-> PARTIAL are already in place.

This allows reuse of blocks, easy patching, and edition management.

For migrating FiveWin applications to the web, I see this patcher/preprocessor workflow as a better path than trying to emulate OOP desktop windows. It is easier to understand, faster to debug, and avoids unnecessary overhead.

Structure and reuse come from conventions and blocks, not from class hierarchies. Testing is simpler because everything stays deterministic: same input → same output.

The goal would be a community-driven XBROWSE Web Clone: a lightweight grid with keyboard navigation, scrolling, filtering, and editing – with as few external libraries as possible, but in the spirit of FiveWin and Clipper. This way nobody has to reinvent the basics, and the community can build on a shared foundation.

Best regards,

Otto

The reality is that many developers refuse migrations - from Clipper to C++, DBF to SQL, or desktop to web.
That's exactly the reason why we created xBasePHP, to help those people who want to keep their existing code working while
gaining modern capabilities.

Your preprocessor approach is smart for pure migration. Our approach is different - we preserve the original OOP structure, so developers can keep their existing codebase and Desktop software unchanged while transparently gaining web deployment. The harbour project was ahead of it's time, keeping OPP Clipper legacy, even before than PHP (version 5 introduced it ).

Check out our infrastructure overview: (A simple copy and paste, automatized, lol)

https://xbasephp.com/infrastructure_chart_1080p.php

Both approaches serve the community, some approaches are free, some need to recover investment, but either free or paid, the community needs this now - yours for clean migration, ours for zero-change preservation. The Clipper/xHarbour/FiveWin community needs all the help it can get! and people like Antonio Linares and the entire x/Harbour developer community deserve recognizance, Geniuses ahead of their time, that sacrificed time, family and health to give us this jewel... that instead of let Clipper die, has reborn in x/Harbour, from the ashes like a Phoenix bird....

So in other words, I don't want to compete with x/Harbour, i want to make sure a single base code is compatible, so it can be compiled, or mounted in a web server or localhost, keeping a single version..... of course we have implemented hundreds of other capabilities, that can be used if wanted, but is not our main goal... we want 100% compatibility, to expand, not compete.. And believe me, i can introduce 100 new features as fast as I produce one of the easiest functions in xharbour.... like Antonio said, is almost impossible, but so it was Project Harbour....

Hello Bayron,

I really enjoyed reading your post – it shows how much passion and vision you put into xBasePHP.

It makes me happy to see how deeply you already master the use of AI in this field – that’s truly inspiring!

I wish you lots of success with your project, and I’m sure it will bring great value to the whole community.

Keep going, Bayron – your work gives energy and hope to all of us who care about keeping xBase/Clipper alive.

Best regards,

Otto

Thanks Otto, it means a lot coming from someone I admire so much,,,, I saw your first tests, and I got overwhelmed by your achievements and efforts, that inspired me as well, to not to wait, and do something about it.... I waited 10 years, then about 5 years ago, I starting testing, purchasing software's, frameworks, this that... testing on translating to different legacy development languages, Cobol, clipper, Delphi. etc....

So its been a need since about 20yrs ago, I been stress testing, solving millions of failed approaches, and then, still going to be another million I did not think of at Open day, and everyone is going to forget what it takes, and bitch about it... but we'll see... plus I'm extending entirely for lots of other stuff.... Crypto, Stock market analysis, etc... so you can create your own stocks wall at home, or use stock movement analytics, Financial banking integration, etc....

I resolved how to integrate AI's and use other AI's as agents, TRADE SECRET for us, but can be investigated in the web.... Got access to the right hardware, borrowed, but still very expensive... they don't charge me for use, but for stuff like energy use, Server Cooling, internet share, etc... but I'm building it and hope when I finish I wont need the infrastructure any more, and will be viable... The product will be light, but the infrastructure in software is already a few terabytes and growing, since is not just the functions, but internet security, logging, error handling, integrations, etc...

So you can see the extension, I'm preparing 2.5 million stress tests for the weekend (That's why you need the equipment infrastructure, but that helped me, because in the process, we produced software that will be multithread, multitask... ). Ill be a long long weekend.... and as I said, the one test we didn't do, is the one that will fail....

Hello Bayron,
Thanks so much for sharing your work here — it really matters. Beyond the “TRADE SECRET for us” parts (which is fair and right), the moral support is huge. Even critical voices help; the worst outcome is when good work isn’t seen or registered at all.

On my side: I run a solid server and a local AI setup, but spent most of the summer refactoring — removing Bootstrap/jQuery/DataTables and stripping table logic from my app. I also tried php4dbf; access was fine and fast, but I didn’t complete CDX handling, so I switched to a HARBOUR/FiveWin microservice. Right now it feels stable and promising.

Can I ask you something? Years ago I built “AUTOCODE for FiveWin” to parse fivewin.ch. Most of it works, but I never finished resource handling and the REDEFINE commands. Do you think your AI approach could help improve that source and add REDEFINE support without a huge effort? If yes, I can post the current code. The idea: getting Harbour/FiveWin code from AI is super practical — you can validate syntax quickly even before compiling.

Appreciate your persistence and the scale of what you’re attempting — stress tests, integrations, the whole stack. Wishing you tons of success. If there’s a small part I can help test or validate, count me in.

Best,
Otto

Please search AUTOCODE in the forum or start here:

How Autocode Works. Here are the main points of how the process works:

**Conversion of preprocessor directives into source code:**

The Harbour compiler handles this task. It translates `#xcommand` directives such as `@ <nRow>, <nCol> BUTTONBMP ...` into executable source code (e.g., `TButtonBmp():New(...)`).
This happens automatically through the PPO file (Preprocessor Output), which shows the resulting source code.

**Reverse translation through Autocode:**

Autocode reads the generated source code from the PPO file.
It converts this source code back into the original directive (e.g., `@ <nRow>, <nCol> BUTTONBMP ...`), extracting the parameter and variable values.
For this, the mapping structure (`M->acd5_params`) is used, which defines how the individual parameters are interpreted and represented in the code.

---

I think it is now possible to automatically generate the mapping blocks using ChatGPT. At least with the blocks I tested, it works.
It would now also be possible to create mapping blocks for the redefine blocks.

However, since there is no interest in Autocode, and I do not work with resources myself and therefore do not need it, mappings are not included in the current `ACD5.ch`.



https://forums.fivetechsupport.com/viewtopic.php?p=276698&hilit=autocode#p276698

Dear Byron,

Could you convert FWH\samples\fivedbu.prg using your system ?

That would be a very interesting test :idea:

Hello Bayron,
That would also be very interesting for me, to compare the source of the PHP-developed version with the version that the AI PHP creates from fivedbu: https://forums.fivetechsupport.com/viewtopic.php?p=280394&hilit=xbrowse+web#p280394

best regards,
Otto

I have only implemented basic commands for transcription, and i need to run a couple million tests on the code but i will try to implement the commands that FiveDBU has to try to make a test,,,,,, ill need a few days.... I'm working in 4 different projects right now, all of similar importance and magnitude... but that will be not that hard to accommodate soon..., I want to run it through the assembly line, it will be a nice test....

Antonio Linares wrote: Dear Byron,

Could you convert FWH\samples\fivedbu.prg using your system ?

That would be a very interesting test :idea:

So back to your request,
we have hit walls left and right, but not wanting to just create a system that recreated only FiveDBU, we have continued creating the Neuro-Network.

We hit walls at... what if the user modified some of the classes, what about .rc or .res files, what about .c and .cpp code, this, that, and that,,,,,....
So we created our own native .res de-compiler, no need for Resource Hacker or anything else... and trying to avoid competition, also we will require the user to provide their own fivewin and harbour/xharbour code... that pushed us back a lot...

We also decided on mobility as well, so we will start with web server based, and as soon as it is stable,,, move to a full .exe compiler... no need of web server to produce the code....

So our intention is to force the user to purchase FiveWin if it doesnt have it, or be limited to write programs for Harbour/xHarbour... and/or the free FiveWin version.


https://www.tiktok.com/t/ZP8A22Em4/

Antonio Linares wrote: Dear Byron,

Could you convert FWH\samples\fivedbu.prg using your system ?

That would be a very interesting test :idea:

In the effort of not just creating the translated code for the app, we have hit many walls, but now, we are very close, we got to the point of creating the window, the reserved space for the drop-menu, and the button-ribbon can already be seen, the problem was that every function, every element, every detail, database management, construction, inheritance, etc... was automatically built from the .prg code, has delayed us, but we are not afraid of trying to see above the wall,,,, thanks for your support.....

Antonio Linares wrote: Dear Byron,

Could you convert FWH\samples\fivedbu.prg using your system ?

That would be a very interesting test :idea:

Already in the works.....

Dear Byron,

What GPU hardware do you have access to ?

Wondering if you have enough hardware to run this:
https://github.com/karpathy/nanochat

https://deepwiki.com/karpathy/nanochat

Right now I'm back to basics for a while,,,, my friend did not really had authorization to give me access to his company hardware..., so i lost that, but i learned that i basically had probably compared access to like 2 RTX3090's processing power, according to my friend, since i didn't had full access.

Virtual use of a H100 probably goes for about $700.00 a month,,, I'm thinking about it,,, but still what you are asking needs 8 H100's... I'm building a couple of houses now, so i can't spare the $6,500.00 for a new rig for my self with a RTX5090,(with propper configuration it will work there, slow, but it will work),,, so I'm thinking seriously about the virtual hardware in the mean time..... projects do not work the way we expect them to work... but if there is a will, there is a way... I just have to make the monthly payment worth and feasible... especially now that Gemini suicided on me, after it could not do several tasks, and now refuses to reinstall..... Long Story....Funny story,, i just told Gemini that it should sell tortillas instead of coding,,,, it did not take it well....LOL....

Any way, I'm sorry i can't do it for now... it says in the repo that you will get a kindergarten kid after training, so it could have been enlightening.... just what we need, to get out of this life of boredom...

Also for what I'm doing now,,,, it looks like my current computer can handle it even without GPU, just using the pay services for AI...Slower but workable...

Dear Byron,

I found this cloud servers company:
https://www.hyperstack.cloud/

I hired a H100-80G-PCIe-NVLink x8 for 15,60 USD x hour. Just four hours are required for the training.
Anyhow I am testing it now with cheaper machines also just in case I may need more and more trainings.

It is just for learning purposes :) I have a RTX3060 eGPU on my iMac Pro but it so small for this code :?