Nicolas Martyanoff

Contrarian software engineer. Hire me to solve your technical problems.

Public Key

npub1dkg70vzk4vvfsc5cup9gkfu2tcrs5unysl7npfvegf78nkuswr5skh54dt

NIP-05 Address

galdor_at_fosstodon.org@mostr.pub

Profile Code

nprofile1qqsxmy08kpt2kxycv2vwqj5ty799upc2wfjg0lfs5kv5ylremwg8p6g0dvn2l

Author Public Key

npub1dkg70vzk4vvfsc5cup9gkfu2tcrs5unysl7npfvegf78nkuswr5skh54dt

Show more details

Last Notes

2025-03-25 22:18:49

I often hear people arguing that they are right because the topic of the conversation is something they do professionally. But anyone working anywhere knows that in all workplaces skill distribution is roughly gaussian.
"It's my job, I know what I'm talking about" is never a good argument.

2025-03-23 23:43:50

- reply

@nprofile…qs0v No, you just made me realize that I messed up my example. The “http” block is of course a direct child of “server”, not part of a listener.
Thank you for noticing !

2025-03-23 17:06:48

I just added the simplest TLS setting ever to Boulevard: add "tls redirect", and all HTTP requests are automatically redirected to the first TLS listener. Bonus, add "hsts true" and it will do the right thing.
If you don't think it's useful, this is a comparison with NGINX:
https://cdn.fosstodon.org/media_attachments/files/114/212/838/157/765/479/original/a990a8e302d00de9.png

2025-03-23 17:03:07

I just added the simplest TLS setting ever to Boulevard: add "tls redirect", and all HTTP requests are automatically redirected to the first TLS listener. Bonus, add "hsts true", and it will do the right thing.
If you don't think it's useful, this is a comparison with NGINX:
https://cdn.fosstodon.org/media_attachments/files/114/212/823/096/562/346/original/9cc820111775230f.png

2025-03-19 21:12:42

If your lead engineer is saying "yes" more often than "no", you're cooked.

2025-03-19 12:55:47

- reply

@nprofile…nwzt the trick is to realize that GitHub lets you block people. No point in arguing in these situations, just block them: no more issues, no more PRs.

2025-03-16 15:54:06

- reply

@nprofile…utqm This changes is announced for 31.1, and we only 30.1 3 weeks ago. ¯\_(ツ)_/¯

2025-03-16 15:41:13

I have updated Boulevard to make HTTP matching simpler. The BCL format supports symbols and string sigils; this is the result: one less indentation level and less setting types.
This is why custom configuration formats are worth it: you can't do that with YAML or TOML.
https://cdn.fosstodon.org/media_attachments/files/114/172/865/251/462/259/original/8b576e46b94d90e6.png

2025-03-15 23:50:38

- reply

@nprofile…utqm I honestly don’t know what to tell you, I spend forever trying to get Font Lock keywords to work with faces. Turns out the documentation is clear, you need a variable whose value is the name of the face. Fun fact, this is the case for all Font Lock default faces.

2025-03-15 22:34:14

- reply

@nprofile…qs0v I must confess than I'm used to having C-k bound to EVAL-BUFFER, same binding as the one ine SLIME for Common Lisp.
Having to move to the end of each expression to evaluate them is asinine.

2025-03-15 22:33:26

TIL that to use a face (e.g. created with DEFFACE) in FONT-LOCK-KEYWORDS (#Emacs), you need to declare a binding whose value is the name of the face. So for a face named FOO-FACE, just add (DEFVAR FOO-FACE 'FOO-FACE). This happens because Font Lock expects faces in FONT-LOCK-KEYWORDS lists to be symbol whose value is the face to use.
I would love to know why Font Lock cannot realize that the symbol is the name of a face and use it directly.

2025-03-15 19:37:08

- reply

@nprofile…qs0v EVAL-LAST-SEXP and EVAL-BUFFER behave differently. I stopped expecting consistency from Emacs a very long time ago.

2025-03-15 18:18:31

- reply

@nprofile…x0kw DEFCUSTOM is for values customizable by the user. In modules, you have a lot of values which are not supposed to be customizable (e.g. regexps used by font-locks) but must be easy to redefine when you're working on the module.

2025-03-15 17:28:47

Let us add DEFAPARAMETER to the long list of features #Emacs Lisp is missing compared to #CommonLisp. ELisp only has DEFVAR, which does not modify the binding if it already exists. Perfect to keep a state. But for a lot of top-level values, you want DEFPARAMETER, which creates or updates a binding.
Module development with DEFVAR only is really frustrating, forced to manually call SETQ each time you want to change a value.

2025-03-10 10:15:51

You'd be surprised how many $100M+ companies publish software affected by known security vulnerabilities and will not release updated versions for weeks or even months because they simply don't care.
Then people ask my why I always encourage companies to invest in tech and build their own platforms: when problems arise, you can fix them instantly.

2025-03-08 18:32:01

Generating FreeBSD packages is usually done with ports. OK if you're working as part of the port tree, not so much if you just want to provide a package.
So I've updated my Fpkg to make it easy to build packages. Compared to the standard Freebsd pkg program, it automatically generate file lists and hashes, makes it easy to specify file owners, groups and permissions, and automatically generates user and group creation scripts.
https://github.com/galdor/fpkg
https://cdn.fosstodon.org/media_attachments/files/114/128/238/297/876/978/original/24540bfe17239769.png

2025-03-08 17:53:50

- reply

@nprofile…qfh3 @nprofile…c4zu I'm curious about your experience in other languages.

2025-03-06 21:32:25

Yes #Python is a deeply unimpressive language, but in my experience it's tolerable for simple tasks as long as it's used correctly.
Most Python code out there is terrible because it's written by less experienced developers who don't read language and tool documentations and copy/paste what they find online (content itself written by junior developers), endlessly amplifying the problem.
This is the curse of languages perceived as easy for beginners.

2025-03-05 11:08:39

Small but useful improvement in SBCL 2.5.2, there is now a USE-VALUE restart when you call a structure constructor with a value having the wrong type. If you're using the REPL a lot, it's one of these small things that helps. Thanks!
#CommonLisp

2025-02-28 20:20:40

The more people work on something at the same time, the more quality will gravitate towards mediocrity. Unavoidable since competence follows a gaussian distribution.

2025-02-25 11:17:58

Today I learned that #PostgreSQL does not support row locks (FOR UPDATE) with SELECT DISTINCT. Not the end of the world, I can filter client side in this case, but I would love to know what led to this limitation.

2025-02-23 15:17:23

- reply

@nprofile…5azp @nprofile…klck @nprofile…8f02
Good books are always useful (I still recommend Practical Common Lisp), but let's not kid ourselves, this is 2025: people new to programming do not start by buying a book.
The problems of the CL ecosystem are well known so I won't rehash them. If you want more CL in the world, go use it and talk about what you've built as publicly as possible.
HN running on SBCL is the latest example I remember, and it's a good one.

2025-02-21 16:58:57

- reply

@nprofile…8f02 What would you like to see in a Common Lisp book that has not been covered by existing ones?

2025-02-21 15:23:56

- reply

@nprofile…8w9a It's not implied, you are just interpreting it that way.
And 0 and 0.0 are two different representations for the same mathematical value. All languages make this distinction, it's nothing unexpected.

2025-02-21 14:01:08

- reply

@nprofile…8w9a There's nothing in RFC 8259 that defines JSON numbers as floats. It mentions that software _can_ use floats and how it would affect interoperability, but number representation remains implementation defined.
And parsers in other languages handle it perfectly. JSON.parse in Ruby will give you an integer for 42 and a float for 42.0, as it should be. Same thing for json.loads in Python. Same thing with the json Racket package. Everywhere really.

2025-02-21 13:55:27

The standard Go HTTP client follows redirections by default. You can disable it with a custom CheckRedirect function in the client. But you cannot disable it for a single request, it's all or nothing. Terrible design.
A workaround is to use Client.Transport.RoundTrip() which will bypass the whole redirection logic of Client.Do(), but if you do so you will also miss features such as request cancellation.

2025-02-19 12:44:51

Go's encoding/json parser is fundamentally broken: it assumes that all numbers are floats (a very common mistake about JSON), meaning that if you decode without a concrete type and re-encode, integers become floats (e.g. {"a":42} -> {"a":42.0}).
The UseNumber() option helps with that, but now you have to deal with a custom json.Number type in all data processing functions.
All they had to do was to check for a '.' character in the token during parsing to identify the type of the number.

2025-02-19 09:13:40

TIL that Hacker News runs on SBCL using Clarc, an Arc implementation written in CL (instead of MzScheme). Something funny to keep in mind next time someone tells you the Common Lisp is dead :)

2025-02-18 11:26:00

- reply

@nprofile…8f02 If you're new to Git, yes absolutely. There's a moment when you understand how it works, and suddenly using it become second nature.

2025-02-18 11:16:03

5min in the Git history of a developer will tell you everything you need to know about their care and attention to details.
If the history is a mess, there's no need to dig further, and the conclusion is not good.

2025-02-16 14:14:44

I just finished a first version of the index view to show the content of directories with Boulevard. Useful when serving simple files.
It does not have to be complicated: add one single setting to your HTTP handler and let Boulevard deal with content negotiation. You get a nice view either in text, HTML for web browsers and even JSON because why not.
https://cdn.fosstodon.org/media_attachments/files/114/013/980/741/610/413/original/50bd03e924881461.png

2025-02-16 09:39:37

In Go, os.ReadDir cannot read entries incrementally and calls lstat on each entry. Terrible interface if you need to handle arbitrarily large directories. A puzzling choice when POSIX already have the correct API (opendir/readdir_r is incremental and fast).
A tolerable alternative is filepath.WalkDir (Go 1.16+): no more lstat, but it still forces you to read all files in each directory.
Seriously bad design.

2025-02-10 14:30:36

InfluxDB 3 has two versions, Core (open source) and Enterprise. The Core version is *really* OSS, MIT/Apache licensed. Perfect example that yes, you can be a company, make money, and still use Open Source licenses.
Bonus: unlike Prometheus, Influx gives me correct percentiles without having to manually define buckets for every single metric.

2025-02-09 13:51:27

Writing an HTTP server / reverse proxy is easy. Make it actually useful for all kinds of use cases is hard; so many small features.
For example people usually expect to be able to customize the response when a file cannot be served.
https://cdn.fosstodon.org/media_attachments/files/113/974/253/004/667/558/original/30686c87a282d9bf.png

2025-02-09 11:57:40

If you're writing a socket and get EPIPE (in Go that would be an op.SyscallError value with Err set to syscall.EPIPE), the connection is down. When that happens for FastCGI connections in Boulevard, I try to reconnect and write again (once of course). This is a simple way to deal with FastCGI servers restarting without clients being affected.

2025-02-07 16:22:49

- reply

@nprofile…29fa Absolutely. I still do not understand why the official answer is not simply "the kernel is written in C, fork it if you don't like it".

2025-02-07 14:17:40

Looking at the open source software ecosystem, I'm astonished that so many developers believe it's their god-given right to have their contributions added to projects.
It get it, it can be frustrating, but unless you *own* the project, why would you think you have any right to decide anything? Discussions on the LKML are just insane.

2025-01-29 16:57:53

The success of DeepSeek, so much more resource efficient than other models is a lesson: if you take a small team of highly competent engineers and micro optimize the hardware at your disposal, you'll easily beat large companies lazily relying on magical high level frameworks.

2025-01-24 12:12:09

Obsessing over database schema rollbacks is a bad idea. They may seem useful on small systems, but the real world is distributed: you can rollback the schema but not the operations that affected other systems, so you end up with an inconsistent system.
Forget rollbacks. Make subsystems easy to quickly disable if they start acting out (e.g. feature flags), and learn to fix my moving forward.

2025-01-17 19:31:56

The Go representation for HTTP headers is incorrect: fields are stored as a hash table, meaning you lose their original order. Most of the time it does not matter, but sometimes threat detection software rely on it (even though they should not).
I'll wait for someone to complain about that for my Boulevard reverse proxy, but if it happens I'll have to parse/serialize headers myself.

2025-01-16 19:27:31

- reply

@nprofile…8f02 There may be a point where "outdated" becomes "vintage" and it's good again :)

2025-01-16 19:14:38

If you're a tech professional and the email address on your resume is gmail dot com (or worse outlook dot com), people will notice. And no it does not look good.

2025-01-16 16:20:27

- reply

@nprofile…29fa I have not seen anyone doing real structural edition using Treesitter beyond one or two examples. I'd love to hear about it if someone is actually doing it though.

2025-01-16 14:03:19

I love Lisp and S-expressions, but I also understand why so many hate the "))))" syndrome with a passion. I'm not affected because I use Paredit mode in Emacs but if you do not have it, keeping parenthesis balanced is painful.
Blame the entire programming world for making structural editing the exception and not the norm.

2025-01-16 11:19:22

- reply

@nprofile…cml5 I'm not sure what you expect. If you're telling Straight to use the main branch (i.e. not using another branch you considered stable or a specific tag), then it will happily do so. Straight does not assume anything: you ask it to use the main branch, it uses it.

2025-01-15 19:05:03

- reply

@nprofile…vatg @nprofile…euxp
Almost everyone has standardized around CommonMark.

2025-01-15 17:50:24

- reply

@nprofile…fkdz There's an entire new generation of tech users who don't care about it because they grew up in a world where big tech surveillance is ubiquitous.

2025-01-15 17:28:56

I thought that Go telemetry was supposed to be opt-in, but I just saw "gopls ** telemetry **" processes running on my machine. Had to run "go telemetry off" (which writes ~/.config/go/telemetry) and restart gopls to get rid of them.
Scary that this is widely accepted.

2025-01-15 13:20:40

A job interview is not just to evaluate your skills, it also is to see how you interact with people and get a feel of what working with you would be.
Aggressively "flipping the script" and trying to show how smart you are may work for the skill part, but will probably make you look insufferable. Tread carefully.

2025-01-14 17:43:13

- reply

@nprofile…tvru I imagine that one is locale-aware while the other is not. I'm really surprised that Float.parse returns 1.0 for "1,23", this is really incorrect. The corresponding Erlang function (string:to_float) correctly returns {error,no_float}.

2025-01-14 15:16:42

- reply

@nprofile…dcnr it’s not a bad book, but the author has a way to write Common Lisp which is quite different from what what you’ll see in the standard or in CLtL2 for example. Something to keep in mind.

2025-01-14 14:09:06

- reply

@nprofile…n8jx It probably applies to all projects in all industries.

2025-01-14 13:45:17

Most software projects succeed not because they have good management and processes but because they are staffed with the right people doing their best to get it done while working around stupid rules.

2025-01-12 01:33:11

The new BCL format for Boulevard has interesting upsides. For example in HTTP handlers you can define the response with just a status code:
reply 200
Or with a status code and body:
reply 200 "Hello!\n"
Or with a full block containing all possible parameters (optional of course):
reply {
status 200
header "Content-Type" "application/json"
header "Content-Language" "en-US"
body "{\"status\": \"ok\"}"
}
Configuration should be easy to read and write.

2025-01-08 21:58:02

- reply

@nprofile…l9v9 Thank you so much :) It always feels good knowing your work helped someone. Enjoy!

2025-01-08 15:30:10

The old experimental Go maps package was very useful. Infortunately since 1.23 the new stable maps package only returns iterators, so every time you need a slice you have to wrap the call with slices.Collect.
I understand the appeal of iterators for optimization (less intermediary allocations), but I'm not a fan of the growing complexity of the langage (e.g. "func EqualFunc[M1 ~map[K]V1, M2 ~map[K]V2, K comparable, V1, V2 any](m1 M1, m2 M2, eq func(V1, V2) bool) bool", not it's not a joke).

2025-01-07 12:42:23

- reply

@nprofile…2sk2 Unbound means there is no value bound to the slot. This does not stop the accessor from binding a value to the slot.
Much less scary than "no applicable method", which is quite clear but can be annoying when you thought you had defined the right method but removed it interactively from some reason and cannot find it again in your source files.

2025-01-06 12:27:08

- reply

@nprofile…qs0v For complex projects, you always end up building your own framework based on your precise needs.
I learned to work like that for system development during my first job. I've been sticking with it for my entire career, it is so much more comfortable.

2025-01-06 11:42:09

- reply

@nprofile…6r8y DSLs in htmx? Where?

2025-01-06 10:40:50

I do not have anything against htmx, it certainly is miles ahead from React/Vue/Angular & co. But I still think its main value is to get web developers on the path to the realization that no, they do not need a framework.

2025-01-05 12:09:45

UNIX sockets are underestimated. They can be used for connections between programs on a local machine (e.g. a CLI program for a server, or a controller and its workers) the same way as a TCP socket, but with the advantage of being actual files with permissions for access control.
The only downside is that some languages/runtimes do not automatically delete
them on close, which is annoying the next time the server starts up.

2025-01-04 12:28:02

Yeah sure, spend hours arguing in favour of HTTP/2 and HTTP/3 when all requests have to go through multiple layers of ridiculous Cloud infrastructure, are handled by slow web application written in Javascript, Ruby or Python, and have to hit one or more databases anyway.
HTTP 1.1 with keep-alive is good enough for the vast majority of use cases. If it's too slow you need a real transport protocol, not HTTP/3.

2025-01-03 08:28:41

- reply

@nprofile…de2h I never used it myself, but you may want to have a look at the dispatch table and deal with that in a more generic way.
But to be honest, you should not use the printed representation of the object as a way to easily dump all fields. This is why DESCRIBE and INSPECT are for.

2025-01-03 08:22:03

- reply

@nprofile…de2h Only for each class for whom you want a custom printed representation. What do you mean for "& field"?

2025-01-03 00:16:40

- reply

@nprofile…zj6d afaik you cannot. I keep my pinned Mastodon tab muted in Firefox. Really annoying (and unfortunately only one entry in the long list of annoying things with the Mastodon web app).

2025-01-02 16:04:09

- reply

@nprofile…8vnl @nprofile…klwf @nprofile…zzgr
FYI your link yields the following error:
wasm streaming compile failed: TypeError: WebAssembly: Response has unsupported MIME type 'application/octet-stream' expected 'application/wasm'
falling back to ArrayBuffer instantiation

2024-12-30 16:15:43

Still working on configuration validation. Most softwares treat configuration errors as an afterthought: print a terse error message and exit.
I'm investing time to do it the right way for Boulevard and for other projects. Worth it.
https://cdn.fosstodon.org/media_attachments/files/113/742/662/835/137/347/original/717c0ee9dd72aa64.png

2024-12-30 15:08:59

- reply

@nprofile…xctk I see the error in the source view (C-u).

2024-12-30 14:40:17

- reply

@nprofile…xctk I love the clean, minimalist design! Just note that the light gray dates on white background are borderline unreadable :)
I also believe there is an error in the generated markup, <span class="displayBlock"/> is not correct in HTML 5 (Firefox displays an error). Not 100% sure though, I'm no frontend developer.

2024-12-29 17:25:20

- reply

@nprofile…qs0v You already see diminishing returns past 6h a day for engineering tasks.

2024-12-29 16:22:06

The only answer to an engineer boasting about working 80h a week is "no you're not". Because no, they are not actually working 7am-9pm 6 days a week. And if they are, they only produce garbage until they burnout in a spectacular way (I saw a senior dev burnout for much less, it was not pretty).
But sure, continue glorifying a macho culture that believes that more work == better results, smart companies will just scoop engineers fed up with your bullshit.

2024-12-28 12:48:40

- reply

@nprofile…579c It's probably more similar to a script language yes.

2024-12-27 18:46:22

- reply

@nprofile…04rq FYI:
> there’s only cl-map, but it requires another argument and using 'cl at runtime is frowned upon.
Nowadays it's cl-lib, and no using it not "frowned upon". It just happens that RMS and some other Emacs developers really hate Common Lisp. But cl-lib is an integral part of Emacs, and it's much more consistent and well thought (because CL) than most other Emacs base modules.

2024-12-27 18:06:03

- reply

@nprofile…579c I have not implemented the access layer yet, but yes it would be based on the verb (entry name): if it's "reply", there must be either one argument (the status code) or two (the status code and the body text).

2024-12-27 16:08:23

- reply

@nprofile…579c In YAML, the status code is an optional entry that defaults to 200. With the new configuration, I cannot really have the same thing, it's either "reply <status>" or "reply <status> <body>".
I can also support a short version (single entry "reply <status> <body>") and a detailed one:
reply {
status <status>
reply <body>
}
In which case status would be optional.
Language design is hard.

2024-12-27 15:01:24

So this is what an example Boulevard configuration would look with the new format on the left compared to YAML on the right.
Obviously it's longer, but it is more readable to me, especially for nested handlers. I'm not 100% convinced though, maybe I'm missing something.
What do you think?
https://cdn.fosstodon.org/media_attachments/files/113/725/386/362/197/728/original/c9f20d79eae1a849.png

2024-12-27 14:33:46

- reply

@nprofile…50n2 Berkeley Mono 2. 100% worth it.

2024-12-26 18:09:19

#Emacs highlights text using font-lock-mode. Keywords highlighting is defined using font-lock-keywords. When set to a list of symbols, it defines multiple levels. You can select the level on a per-mode basis using font-lock-maximum-decoration.
Personally I prefer using less colors, but some users may have different preferences. So I now support 3 levels for bcl-mode At the highest levels, even entry names and block types are highlighted.
https://cdn.fosstodon.org/media_attachments/files/113/720/462/581/079/676/original/77918a33fefee6de.png

2024-12-26 14:27:44

- reply

[2/2]
Then I ran some tests and realized NGINX has no problem with multiple settings. Try it with multiple "return" statements, the first one is used (would have prefered the last one though).
If you consider entries as statements instructing the software to do something and not settings, it starts to make sense. So let's go with the simpler option.

2024-12-26 14:27:37

[1/2]
I've hesitated a lot between two options.
The left option is simpler, but requires allowing duplicate entries. The right option preserves unicit, but is more complex and relies on keys being potentially strings and not just symbols.
https://cdn.fosstodon.org/media_attachments/files/113/719/591/083/049/566/original/a3fcdb55582cb944.png

2024-12-24 15:51:44

Designing and implementing a new language for configuration files is one thing, but there is no way I'm using it without an Emacs mode. So I wrote one!
That was not easy. Inheriting from prog-mode helps a lot of course. Then defining a syntax table, and suddenly font-lock magically highlights comments and character strings. Indentation (indent-line-function) was painful, but I got it working for the canonical representation (no inline blocks, sorry not sorry).
https://github.com/galdor/bcl-mode

2024-12-24 13:49:05

- reply

@nprofile…pard Yes, without age correlation it does not mean much.
From what I have seen these last years with new generations of developers —let us say those under 30— Emacs simply does not exist.
They may have heard of it, but it's perceived as some kind of relic used only by old grey beards.
Funny thing is that those who see experienced Emacs users work are absolutely amazed by what is possible. But you cannot compete with the massive marketing campaign of VS Code (and one should not try).

2024-12-23 16:10:11

- reply

@nprofile…jqwa Slower than what? And in which Lisp dialect? At least in Common Lisp, quasiquoting has no impact on execution time, all the magic happens at read time (and yes I'm aware that the devil is in the details).

2024-12-21 16:55:03

- reply

@nprofile…x6hw I do to, it allows for more compact representations. It also means that for an element, you can clearly lookup a child block ("give me the listener named 'foo' for this server").
And of course, supporting named blocks does not mean you cannot have unnamed ones.

2024-12-21 16:39:43

- reply

[2/2]
I also wrote a pretty printer, but it will not be usable until I parse comments and store them in the AST instead of discarding them. Something for another day.
Of course I need tests, a specification, an Emacs mode, but it can wait. Now I want to switch Boulevard from YAML to BCL and see how I feel about it.

2024-12-21 16:39:35

[1/2]
I have finished a minimal Go module for my BCL configuration language (see my GitHub account), BCL (for Block-based Configuration Language). I still have to figure out if I want to support named blocks, for example:
listener "http" {
address ":80"
}
listener "https" {
address ":443"
}
instead of:
listener {
name "http"
address ":80"
}
listener {
name "https"
address ":443"
}

2024-12-15 18:56:39

I wrote a tokenizer and parser for my configuration language. The important part is making sure users get clear and accurate errors when they make mistakes.
This is starting to look good:
https://cdn.fosstodon.org/media_attachments/files/113/658/361/525/871/604/original/3dedac81336a9f7b.png

2024-12-15 11:27:28

People leave bosses, not companies. They just don't talk about it publicly because image.
Average tenure in tech companies is ~2y. No comment.

2024-12-12 12:21:11

- reply

@nprofile…qgqt Ruby is a good example yes, you might be right.
Maybe it's just in my head, I would feel strange writing a setting such as:
redirect 301, "https://example.com"
The comma feels unnecessary, out of place.

2024-12-12 11:54:33

- reply

@nprofile…qgqt Ah yes being specific can help, but I hope to use this format for more than one format.
The argument separator is not a bad idea if you know all your arguments are semantically the same (i.e. it's a list).

2024-12-12 11:34:35

With a configuration language, you must decide on how to end lines. With semicolons (e.g. NGINX) it is explicit but more verbose. Without them (e.g. Caddy or HAProxy) you must either forbid multiline statements (HAProxy does that, really annoying) or provide a specific construct.
A backslash character works, as in shell or Ruby. It's not pretty, but people are used to it.
The most important reason is that the vast majority of settings should be single line, so we should optimize for them.
https://cdn.fosstodon.org/media_attachments/files/113/639/636/798/176/264/original/486848fb06defc46.png

2024-12-11 20:04:47

- reply

@nprofile…6r8y Not necessarily. Formats with simple key/values entries and blocks are used in a lot of software.
It just happens to be that they are all custom: no specification, no library. But this can be fixed.

2024-12-11 20:03:29

- reply

@nprofile…yg9s The absence of comments is an immediate deal breaker for configuration files.

2024-12-11 19:42:30

- reply

@nprofile…6r8y I appreciate the link, and I want to learn more about it because it is interesting, but I suspect most hate Haskell about as much as they hate Lisp.

2024-12-11 19:25:44

- reply

And before anyone asks: yes I like s-expressions, but most of the rest of the world hates them, so…

2024-12-11 19:25:02

- reply

@nprofile…skk3 Indentation only is one of the major issues with YAML: it's nice for short, simple documents, but makes longer configuration much harder to read.

2024-12-11 19:16:59

Ok I'm fed up with YAML, it just does not work for large and complex configuration files (e.g. Boulevard: imagine an array of dozens of handlers…).
So I'm playing with a new custom format. Well not so custom, it's very similar to what NGINX has. Easier to read, without deeply nested objects, designed for programs needing more than a couple keys/values.
If I go all the way I'll write a specification and a reference library. Anyone interested?
https://cdn.fosstodon.org/media_attachments/files/113/635/792/349/072/760/original/e643085691c3f7b7.png

2024-12-11 09:26:30

- reply

@nprofile…de2h You must have defined SYM->STR at some point in CL-USER instead of doing so in your MARKLIB package.
Unintern it in CL-USER and you'll be fine.

2024-12-10 13:04:27

With software, learn to be as strict as possible. Then you'll understand when you can afford to be sloppy. If you start with sloppiness, you'll never be able to be strict when it's necessary. And it applies to everything, not just software.

2024-12-09 11:44:59

An important step of your journey as a professional is to realize that guidance may never come. If you find a good mentor, wonderful, but it's rare. The earlier you accept that, the faster you'll get back on tracks.