It's been very interesting to see the recent flood of bug detection and fixes. Hlöðver, how are you discovering all of these issues? Are you using AI or some other tool, test results, or some combination? It's rather amazing to witness, and amazing to see how many hidden bugs exist in the codebase. 

Hi,

yes it's discovered with GPT6 asta, high level.

Yes, the first bug I found by accident while working on another project, then I started searching for similar bugs and then now I have a loop that can discover any category of bug within some constraints I give the agent. It's good to get this scan finished before a major release. Most of these bugs are minor, or require passing a very specific set of parameters to encounter them. It would take me months to discover what the LLM has discovered in last week. And I don't think we have the time required to scan the entire codebase like an LLM can do. So I'd argue, we can focus our creativity with Csound, knowing that it's going to behave as we expect it to.

On Thu, 17 Sept 2026 at 17:20, Dave Seidel <dave.seidel@gmail.com> wrote:

It's been very interesting to see the recent flood of bug detection and fixes. Hlöðver, how are you discovering all of these issues? Are you using AI or some other tool, test results, or some combination? It's rather amazing to witness, and amazing to see how many hidden bugs exist in the codebase. 

yes, it's mostly hardening the codebase. A lot of it is edge cases, particularly when it comes to longstanding code that has been used by many. It's like having the whole world running Csound in all possible ways.

Prof. Victor Lazzarini

Maynooth University

Ireland

On 17 Sep 2026, at 16:45, Hlöðver Sigurðsson <hlolli@gmail.com> wrote:



*Warning*

This email originated from outside of Maynooth University's Mail System. Do not reply, click links or open attachments unless you recognise the sender and know the content is safe.

Hi,

yes it's discovered with GPT6 asta, high level.

Yes, the first bug I found by accident while working on another project, then I started searching for similar bugs and then now I have a loop that can discover any category of bug within some constraints I give the agent. It's good to get this scan finished before a major release. Most of these bugs are minor, or require passing a very specific set of parameters to encounter them. It would take me months to discover what the LLM has discovered in last week. And I don't think we have the time required to scan the entire codebase like an LLM can do. So I'd argue, we can focus our creativity with Csound, knowing that it's going to behave as we expect it to.

On Thu, 17 Sept 2026 at 17:20, Dave Seidel <dave.seidel@gmail.com> wrote:

It's been very interesting to see the recent flood of bug detection and fixes. Hlöðver, how are you discovering all of these issues? Are you using AI or some other tool, test results, or some combination? It's rather amazing to witness, and amazing to see how many hidden bugs exist in the codebase. 

IMO, it's an appropriate use of AI to do the work that no human is likely or willing to do. But it's also true that it would not happen without  Hlöðver's guidance, along with reviews by yourself and other humans. 

On Thu, Sep 17, 2026 at 12:10 PM vlz <viclazzarini@gmail.com> wrote:

yes, it's mostly hardening the codebase. A lot of it is edge cases, particularly when it comes to longstanding code that has been used by many. It's like having the whole world running Csound in all possible ways.

Prof. Victor Lazzarini

Maynooth University

Ireland

On 17 Sep 2026, at 16:45, Hlöðver Sigurðsson <hlolli@gmail.com> wrote:



*Warning*

This email originated from outside of Maynooth University's Mail System. Do not reply, click links or open attachments unless you recognise the sender and know the content is safe.

Hi,

yes it's discovered with GPT6 asta, high level.

Yes, the first bug I found by accident while working on another project, then I started searching for similar bugs and then now I have a loop that can discover any category of bug within some constraints I give the agent. It's good to get this scan finished before a major release. Most of these bugs are minor, or require passing a very specific set of parameters to encounter them. It would take me months to discover what the LLM has discovered in last week. And I don't think we have the time required to scan the entire codebase like an LLM can do. So I'd argue, we can focus our creativity with Csound, knowing that it's going to behave as we expect it to.

On Thu, 17 Sept 2026 at 17:20, Dave Seidel <dave.seidel@gmail.com> wrote:

It's been very interesting to see the recent flood of bug detection and fixes. Hlöðver, how are you discovering all of these issues? Are you using AI or some other tool, test results, or some combination? It's rather amazing to witness, and amazing to see how many hidden bugs exist in the codebase. 

I completely agree with Eduardo. We’re human.

There is a lot to think about here, but right now, in some specific areas, the way we use AI seems almost backwards. We should first think about the 

Cases we want to test and write human readable tests, and only then work on the backend. 

I often find myself rereading tests or fixes generated by AI and struggling to fully understand them. In the long run, I think there’s also a risk of gradually getting used to simply accepting whatever the AI writes.

And of course, this doesn’t apply only to tests. The same concern applies to backend code as well. In my case, I’m still quite slow when writing code, and for now I often prefer to write it myself by hand. 

Of course, AI can help us identify cases we hadn’t thought of, and it can also point us toward optimizations we may have missed or simply didn’t know about. 

But this leads to a broader question: if we start accepting things we don’t fully understand, how can we then delegate to someone else the responsibility of validating or reviewing what has been generated? 

Personally, I’m trying to gradually change my approach. I know I need to learn more about the tools and techniques AI uses for testing, understand them properly, and slowly start incorporating them into my own routine. 

For now, though, a lot of it feels like an unfamiliar language.

And I have to admit that changing this approach is not easy, because there’s always the feeling that if you slow down to understand things properly, you risk falling behind. 

This is just a small reflection, but I think it’s something worth thinking about. 

Of course, none of this takes away from my admiration for the great work hlolli is doing to improve Csound. On that, I completely agree with Victor.

Thank you.

 

Pasquale Mainolfi
Composer and Computer Scientist

Professor of Computer Music
Dep. of New Technologies and Musical Languages, 
"G. Martucci” State Conservatoire of Musica (Salerno, IT).
Professor of Electroacoustic Music Analysis at School of Electronic Music, 
“N. Sala” State Conservatoire of Music (Benevento, IT).



Le informazioni contenute nella presente comunicazione e i relativi allegati possono essere riservate e sono, comunque, destinate esclusivamente ai destinatari sopraindicati. La diffusione, distribuzione e/o copiatura del documento trasmesso da parte di qualsiasi soggetto diverso da destinatario è proibita, sia ai sensi dell’art. 616 c.p., che ai sensi del D.Lgs. n. 196/2003.

The information contained in this communication and its attachments may be confidential and is, in any case, intended solely for the recipients indicated above. The dissemination, distribution, and/or copying of the transmitted document by party other than the intended recipient is prohibited, pursuant to Article 616 of the Penal Code and Legislative Decree no. 196/2003.

Il giorno 18 set 2026, alle ore 05:45, Hlöðver Sigurðsson <hlolli@GMAIL.COM> ha scritto:

In those cases you describe. I can leave an open ticket and leave any ai out of a code that makes it less maintainable.

On Fri 18. Sep 2026 at 01:49, Eduardo Moguillansky <eduardo.moguillansky@gmail.com> wrote:

I am reviewing the PRs which were pointed my way and many are complete rewrites, with unreadable tests. I cannot review most of them and most certainly cannot maintain the new code. 

On Thu, Sep 17, 2026 at 6:20 PM Dave Seidel <dave.seidel@gmail.com> wrote:

IMO, it's an appropriate use of AI to do the work that no human is likely or willing to do. But it's also true that it would not happen without  Hlöðver's guidance, along with reviews by yourself and other humans. 

On Thu, Sep 17, 2026 at 12:10 PM vlz <viclazzarini@gmail.com> wrote:

yes, it's mostly hardening the codebase. A lot of it is edge cases, particularly when it comes to longstanding code that has been used by many. It's like having the whole world running Csound in all possible ways.

Prof. Victor Lazzarini

Maynooth University

Ireland

On 17 Sep 2026, at 16:45, Hlöðver Sigurðsson <hlolli@gmail.com> wrote:



*Warning*

This email originated from outside of Maynooth University's Mail System. Do not reply, click links or open attachments unless you recognise the sender and know the content is safe.

Hi,

yes it's discovered with GPT6 asta, high level.

Yes, the first bug I found by accident while working on another project, then I started searching for similar bugs and then now I have a loop that can discover any category of bug within some constraints I give the agent. It's good to get this scan finished before a major release. Most of these bugs are minor, or require passing a very specific set of parameters to encounter them. It would take me months to discover what the LLM has discovered in last week. And I don't think we have the time required to scan the entire codebase like an LLM can do. So I'd argue, we can focus our creativity with Csound, knowing that it's going to behave as we expect it to.

On Thu, 17 Sept 2026 at 17:20, Dave Seidel <dave.seidel@gmail.com> wrote:

It's been very interesting to see the recent flood of bug detection and fixes. Hlöðver, how are you discovering all of these issues? Are you using AI or some other tool, test results, or some combination? It's rather amazing to witness, and amazing to see how many hidden bugs exist in the codebase. 

The limitation of AI will always be the quality of the question. Thorin and I recently questioned ChatGPT about its ability to develop a knowledge base through questions asked by a community of developers rather than questions asked by one person. The answer from ChatGPT included this : “… ChatGPT now supports shared Projects, where members can contribute chats and files to the same project, and ChatGPT can draw on that shared material when answering future questions inside the project. Shared projects are specifically designed to evolve through sustained contributions from multiple people”. Enlisting the help of ChatGPT to manage shared Projects might lighten the load of the handful of people that keep Csound viable. This could be worth discussing when we meet at Trapani.

On 18 Sep 2026, at 22:22, Victor Lazzarini <000010b17ddd988e-dmarc-request@LISTSERV.HEANET.IE> wrote:

I think that is right, but I guess we'll share that ownership across the dev team anyway. There is a lot of code from contributors who are no longer with us and we just have to deal with it. But as you said we need to ensure any changes can be human maintained as a baseline (even if some of the legacy code does not feel like it to many of us...)

========================================

Prof. Victor Lazzarini

Director

Maynooth University Arts and Humanities Institute

---

From: Csound-developers <CSOUND-DEV@LISTSERV.HEANET.IE> on behalf of Eduardo Moguillansky <eduardo.moguillansky@GMAIL.COM>
Sent: Friday 18 September 2026 12:05
To: CSOUND-DEV@LISTSERV.HEANET.IE <CSOUND-DEV@LISTSERV.HEANET.IE>
Subject: Re: [Csnd-dev] [EXTERNAL] Re: [Csnd-dev] recent bug fixing activity

 

In general it must be clear to the author of a PR that the last hands on a certain piece of code, it they imply more than a minor fix, own that code. It seems temerary to assume such a burden for one particular individual

On Fri, Sep 18, 2026 at 9:40 AM Pasquale Mainolfi <mnlpql@gmail.com> wrote:

I completely agree with Eduardo. We’re human.

There is a lot to think about here, but right now, in some specific areas, the way we use AI seems almost backwards. We should first think about the 

Cases we want to test and write human readable tests, and only then work on the backend. 

I often find myself rereading tests or fixes generated by AI and struggling to fully understand them. In the long run, I think there’s also a risk of gradually getting used to simply accepting whatever the AI writes.

And of course, this doesn’t apply only to tests. The same concern applies to backend code as well. In my case, I’m still quite slow when writing code, and for now I often prefer to write it myself by hand. 

Of course, AI can help us identify cases we hadn’t thought of, and it can also point us toward optimizations we may have missed or simply didn’t know about. 

But this leads to a broader question: if we start accepting things we don’t fully understand, how can we then delegate to someone else the responsibility of validating or reviewing what has been generated? 

Personally, I’m trying to gradually change my approach. I know I need to learn more about the tools and techniques AI uses for testing, understand them properly, and slowly start incorporating them into my own routine. 

For now, though, a lot of it feels like an unfamiliar language.

And I have to admit that changing this approach is not easy, because there’s always the feeling that if you slow down to understand things properly, you risk falling behind. 

This is just a small reflection, but I think it’s something worth thinking about. 

Of course, none of this takes away from my admiration for the great work hlolli is doing to improve Csound. On that, I completely agree with Victor.

Thank you.

 

Pasquale Mainolfi
Composer and Computer Scientist

Professor of Computer Music
Dep. of New Technologies and Musical Languages, 
"G. Martucci” State Conservatoire of Musica (Salerno, IT).
Professor of Electroacoustic Music Analysis at School of Electronic Music, 
“N. Sala” State Conservatoire of Music (Benevento, IT).



Le informazioni contenute nella presente comunicazione e i relativi allegati possono essere riservate e sono, comunque, destinate esclusivamente ai destinatari sopraindicati. La diffusione, distribuzione e/o copiatura del documento trasmesso da parte di qualsiasi soggetto diverso da destinatario è proibita, sia ai sensi dell’art. 616 c.p., che ai sensi del D.Lgs. n. 196/2003.

The information contained in this communication and its attachments may be confidential and is, in any case, intended solely for the recipients indicated above. The dissemination, distribution, and/or copying of the transmitted document by party other than the intended recipient is prohibited, pursuant to Article 616 of the Penal Code and Legislative Decree no. 196/2003.

Il giorno 18 set 2026, alle ore 05:45, Hlöðver Sigurðsson <hlolli@GMAIL.COM> ha scritto:

In those cases you describe. I can leave an open ticket and leave any ai out of a code that makes it less maintainable.

On Fri 18. Sep 2026 at 01:49, Eduardo Moguillansky <eduardo.moguillansky@gmail.com> wrote:

I am reviewing the PRs which were pointed my way and many are complete rewrites, with unreadable tests. I cannot review most of them and most certainly cannot maintain the new code. 

On Thu, Sep 17, 2026 at 6:20 PM Dave Seidel <dave.seidel@gmail.com> wrote:

IMO, it's an appropriate use of AI to do the work that no human is likely or willing to do. But it's also true that it would not happen without  Hlöðver's guidance, along with reviews by yourself and other humans. 

On Thu, Sep 17, 2026 at 12:10 PM vlz <viclazzarini@gmail.com> wrote:

yes, it's mostly hardening the codebase. A lot of it is edge cases, particularly when it comes to longstanding code that has been used by many. It's like having the whole world running Csound in all possible ways.

Prof. Victor Lazzarini

Maynooth University

Ireland

On 17 Sep 2026, at 16:45, Hlöðver Sigurðsson <hlolli@gmail.com> wrote:



WARNING

 

This email originated from outside of Maynooth University's Mail System. Do not reply, click links or open attachments unless you recognise the sender and know the content is safe.

Hi,

yes it's discovered with GPT6 asta, high level.

Yes, the first bug I found by accident while working on another project, then I started searching for similar bugs and then now I have a loop that can discover any category of bug within some constraints I give the agent. It's good to get this scan finished before a major release. Most of these bugs are minor, or require passing a very specific set of parameters to encounter them. It would take me months to discover what the LLM has discovered in last week. And I don't think we have the time required to scan the entire codebase like an LLM can do. So I'd argue, we can focus our creativity with Csound, knowing that it's going to behave as we expect it to.

On Thu, 17 Sept 2026 at 17:20, Dave Seidel <dave.seidel@gmail.com> wrote:

It's been very interesting to see the recent flood of bug detection and fixes. Hlöðver, how are you discovering all of these issues? Are you using AI or some other tool, test results, or some combination? It's rather amazing to witness, and amazing to see how many hidden bugs exist in the codebase. 

The information contained in this email may be confidential and privileged. It is intended only for the addressee(s) stated above. If you are not an addressee any use, dissemination, distribution, publication or copying of the information contained in this email is strictly prohibited. If you have received this email in error, please immediately notify us by email at dataprotection@mu.ie and delete this email from your system.

Please note that Maynooth University is subject to Freedom of Information and Data Protection laws. We may be required to disclose the content of emails under the FOI Act 2014, the Data Protection Act 2018 or GDPR.

Is don seolaí / do na seolaithe thuasluaite amháin an ríomhphost seo. D’fhéadfadh an t-eolas atá ann a bheith rúnda agus faoi phribhléid. Mura seolaí tú, tá cosc iomlán ar aon eolas atá sa ríomhphost seo a úsáid, a scaipeadh, a dháileadh, a fhoilsiú nó a chóipeáil. Má fuair tú an ríomhphost seo trí thimpiste, cuir sin in iúl dúinn láithreach trí ríomhphost a chur chuig  dataprotection@mu.ie  agus scrios an ríomhphost seo ó do chóras.

Tabhair faoi deara go bhfuil Ollscoil Mhá Nuad faoi réir dhlíthe um Shaoráil Faisnéise agus um Chosaint Sonraí. D’fhéadfadh ceangal a bheith orainn ábhar ríomhphoist a nochtadh faoin Acht um Shaoráil Faisnéise 2014, faoin Acht um Chosaint Sonraí 2018 nó faoi GDPR.

Registered charity number 20037130