# -*- eval: (require 'ox-hugo) -*- #+title: Phundrak’s Blog #+author: Lucien “Phundrak” Cartier-Tilet #+hugo_base_dir: ../ #+hugo_section: ./ #+hugo_categories: emacs linux conlanging orgmode #+startup: content * [EN] Open-Sourcing ALYS :ALYS: :PROPERTIES: :EXPORT_FILE_NAME: open-sourcing-alys :EXPORT_DATE: 2021-12-15 :export_hugo_menu: :menu "main" :END: #+TOC: headlines 1 local ** Too Long, Didn’t Read VoxWave no longer exists as a company, ALYS lives on as an open-source project under the GPL-3.0 and the BY-CC-4.0 license. You can find it at the following address: +[[https://labs.phundrak.com/phundrak/ALYS]]+ EDIT: The main repository moved to [[https://labs.phundrak.com/ALYS/ALYS]] and vocal libraries are now separated in different repositories linked from the main one. ** What happened? You might have noticed it, but VoxWave became quite silent over the last months. This is because we at the head of VoxWave chose to close the company, a decision which came in effect in early September 2021. There’s not much else to say. *However*, the good news is the rest still goes on! ALYS as a project is still alive and well! As her creator, I decided to step in and continue its technical support. Since the company no longer exists, and as a free and open-source software supporter, I also decided to open-source ALYS as much as possible. As a result: #+begin_center ALYS for Alter/Ego is now free software as in /free beer/. ALYS for UTAU, including its previously unreleased UTAU prototype, is now free as in /free beer/ and in /freedom/. #+end_center You can find the installer of ALYS for Alter/Ego on the repository linked above as well as a free licence file. Regarding its UTAU version, its prototype is already configured with ~oto.ini~ files, but the source file for its Alter/Ego version are stripped of any configuration. ** What’s New? Therefore, ALYS is now available under three different licences: - The character is now under the free [[https://creativecommons.org/licenses/by-nc/4.0/][CC-BY-NC-4.0 licence]] - The UTAU vocal libraries are now under the [[https://www.gnu.org/licenses/gpl-3.0.en.html][GPL-3.0license ]] ([[https://choosealicense.com/licenses/gpl-3.0/][short readable version]]) - The Alter/Ego is still under a proprietary licence under my name, but it is now available free of charge Basically, this means you can do whatever you wish with the character as long as it is non-commercial, and you credit [[https://www.instagram.com/hsaphirya/][Saphirya]], ALYS’ designer. The UTAU vocal libraries can be used, modified, and redistributed as much as you wish as long as it stays under the GPL-3.0 licence. And you are free to use the Alter/Ego vocal libraries as much as you wish, but you cannot redistribute or modify them. I also decided to release ALYS’ very first, secret, unreleased, unheard French vocal library. It was scrapped no long after recording it due to quality issues and was replaced by its French UTAU prototype people could hear through ALYS’ first songs. It is released more as a way of preserving the fact it existed rather preserving a usable vocal library. (I don’t even remember what it sounds like.) If you have any question, you are free to email me at [[mailto:lucien@phundrak.com][lucien@phundrak.com]] or open an issue on the repository mentioned above. * Conlanging :@conlang: ** TODO Writing my conlanging docs with Emacs :emacs:conlanging: * Development :@dev: ** [FR] Mettre à niveau mes sites org-mode :dev:emacs: :PROPERTIES: :EXPORT_FILE_NAME: mettre-a-nivea-mes-sites-org-mode :EXPORT_DATE: 2022-08-15 :export_hugo_menu: :menu "main" :END: *** Le Problème Cela fait quelques temps que je réfléchis à une nouvelle manière de gérer deux de mes sites web, [[https://conlang.phundrak.com][conlang.phundrak.com]] et [[https://config.phundrak.com][config.phundrak.com]]. Les deux sites sont actuellement générés via un export depuis org-mode (un des nombreux modes d’Emacs) directement vers du HTML. Sauf que l’organisation du fichier HTML de sortie de me plaît pas, et depuis plus de deux ans j’utilise un script rédigé en Dart et compilé vers Javascript pour réorganiser les fichiers. En soit ce ne serait pas trop grave si mes pages web n’étaient pas forcément lourdes. Mais elles le sont! La plus lourde page de mon site de linguistique fait 232Ko (la page francophone sur le Proto-Ñyqy) et celle de mon site de configuration fait 5,5Mo (configuration Emacs) ! Je parle bien de fichiers HTML ! Il faut vraiment que ça change! *** Nouveau Framework pour le front-end À la base je m’étais lancé pour écrire un exporteur personnalisé pour exporter mes fichiers org-mode vers des fichiers JSX qui seraient utilisés par un projet [[https://reactjs.org/][Reac]]t, ou même [[https://nextjs.org/][Next.js]]. Mais j’ai récemment découvert quelque chose qui pourrait être bien plus pratique pour moi : [[https://vuejs.org/][Vue]] et tout particulièrement [[https://v3.nuxtjs.org/][Nuxt]] ! En effet, Nuxt lit le [[https://content.nuxtjs.org/guide/writing/mdc/][MDC]], ou Markdown Components. De fait, il est possible avec MDC et Nuxt d’insérer dans du Markdown des composants Vue soit en blocs soit en inline. Et pour moi, ça change tout ! Je peux maintenant écrire un exporteur minimal qui se chargera simplement d’exporter quelques éléments personnalisés vers des composants Vue, voire même de simples macros org-mode pour exporter les composants inline. Et bien sûr, pour pallier au problème de fichiers HTML trop lourds, il me faudra séparer mes fichiers actuels en plusieurs fichiers, mais cela devrait être plus simple à gérer une fois la transition vers le nouveau framework effectuée. *** Et pour le backend ? Mais ce n’est pas tout : un élément que j’aimerais ajouter à mon site de linguistique serait un dictionnaire entre mes langues construites et d’autres langues, qu’elles soient construites ou non. Ce dictionnaire doit pouvoir être interactif, avec par exemple une recherche, une page par mot, etc. Je ne ferai certainement pas télécharger à mes utilisateurs l’entièreté du dictionnaire à chaque recherche d’un mot dans le dictionnaire, il ne peut donc pas être hébergé avec mon frontend, et j’aurai besoin d’un backend avec une API REST pour gérer les requêtes des visiteurs du site web. Maintenant la question est, quel type de back-end ? Tout d’abord, je vais complexifier un peu le problème : je suis un grand amateur de org-mode. Je pourrais gérer ça via une base de données classique, ajoutant chaque entrée manuellement, mais je vais plutôt essayer de gérer tout ça via org-mode. Les fichiers texte sont plus simples à versionner que des bases de données en un seul fichier binaire. Du coup, il va falloir que je m’écrive un nouvel exporter, mais lequel ? Je pourrais rédiger un exporteur pour mon fichier ~dictionnaire.org~ qui l’exporterait vers un fichier Json qui serait lu ensuite par mon backend qui extraierait et enverrai à mes utilisateurs les informations nécessaires. L’avantage serait de n’avoir quasiment pas besoin de manipuler le Json et d’en envoyer tel quel. Mais l’ouverture et fermeture constante du fichier n’est pas forcément la meilleure des idées, quoi que cela pourrait permettre de remplacer le fichier pendant que le backend tourne. Mais je suis sûr qu’on peut mieux faire. Ma solution suivante était d’utiliser EmacSQL, un paquet Emacs lui permettant d’interagir avec des bases de données SQLite, PostgreSQL et MySQL. Au moins ce serait une véritable base de données, avec seulement un blob binaire à mettre à jour, et ce serait potentiellement plus performant étant donné qu’il n’y aura qu’à ouvrir une fois une connexion avec elle. Mais le problème est maintenant sa mise à jour. Mince… Vient enfin ma troisième solution qui, je pense, sera celle que je vais adopter : utiliser une base de donnée type Firebase. L’idée d’un verrouillage fournisseur ne me plaît pas franchement, donc j’ai décidé d’utiliser une alternative open source et hébergeable : [[https://appwrite.io/][Appwrite]]! Je peux écrire sur une de ses bases de données pendant que mes utilisateurs peuvent la lire, donc la mise à jour n’est pas un problème, et je n’ai rien à mettre en ligne, seulement une série de requêtes à faire. Cependant, un problème reste : comment communiquer avec Appwrite? *** La quête pour un SDK Appwrite pour Emacs Hélas, j’ai beau chercher, il n’existe aucun paquet pour Emacs permettant une communication avec Appwrite. Mais ce n’est pas franchement surprenant : Appwrite n’est pas encore extrêmement répandu, et même Firebase ne dispose pas de paquet pour Emacs. Bien heureusement, Appwrite dispose d’une API REST assez bien documentée, et Emacs est capable de gérer des requêtes nativement via sa bibliothèque ~url~, c’est donc naturellement que j’ai commencé à travailler sur ~appwrite.el~, un SDK Appwrite pour du Emacs Lisp. J’aurais pu utiliser ~request.el~, un paquet assez populaire pour Emacs afin de gérer les requêtes HTTP, mais je ne suis pas grand fan de son workflow et je préfère limiter au maximum le nombre de dépendances dans mes paquets. Ce que ce paquet fait actuellement est une transformation des paramètres nommés que mes fonctions acceptent en un payload Json. Par exemple, ma fonction ~appwrite-stogare-list-buckets~ accepte les mot-clefs ~search~, ~limit~, ~offset~, ~cursor~, ~cursor-direction~ et ~order-type~. Ces arguments sont transformés en du Json via la bibliothèque native d’Emacs afin de donner ceci : #+begin_src js { "search": "my search request", "limit": 30, "offset": 0, "cursor": "", "cursorDirection": "before", "orderType": "ASC", } #+end_src Ce payload Json est enfin envoyé à l’API REST correspondante, en l’occurrence ~/v1/storage/buckets~ comme on peut le voir [[https://appwrite.io/docs/server/storage?sdk=nodejs-default#storageListBuckets][sur la documentation officielle]]. Bien sûr, les éléments optionels ne sont pas nécessairement inclus afin d’éviter à avoir à envoyer trop d’informations. Dans ce cas, tous les éléments du payload sont optionels, ce qui ferait que le ~appwrite.el~ n’enverra que src_js{{}} comme payload à l’API. Pour l’instant, le projet en est encore à ses débuts, mais j’ai commencé à travailler sur le SDK pour Appwrite que vous pouvez trouver sur [[https://github.com/Phundrak/appwrite.el][ce dépôt Github]]. La question maintenant est : comment exporter mon dictionnaire vers Appwrite ? La réponse me semble relativement simple ; je pourrai écrire un exporteur org-mode dépendant de ~appwrite.el~ qui exportera pour chaque mot qu’il rencontrera un payload Json vers mon instance personnelle Appwrite. Et à la différence des exporteurs org-mode habituels, ~ox-appwrite~ n’exportera aucun fichier sur mon système. *** Conclusions Au fur et à mesure de mon analyse du projet et de mes besoins, je me suis rendu compte que j’aurai besoin d’outils plus intelligents que de simples pages HTML exportées automatiquement via Emacs. Ainsi, j’aurai besoin de créer un site web avec Nuxt, profitant ainsi de sa capacité à rendre du Markdown avec du contenu interactif, agissant en tant que frontend pour mon site web. Ce Markdown sera exporté via org-mode à partir de mes fichiers déjà existants, bien qu’à fragmenter afin de réduire la taille des fichiers de sortie. Le backend sera une instance Appwrite que j’hébergerai moi-même sur mes serveurs. Elle sera populée par un exporter org-mode custom via Emacs, ce qui me permettra de continuer à gérer mes dictionnaires et mes langues avec org-mode. Ce projet est vraiment intéressant car cela m’a incité à explorer de nombreuses possibilités et technologies différentes afin de trouver ce qui correspond le mieux à mon besoin, notamment en me rendant compte par exemple que React n’était pas forcément l’outil le plus adapté à ce projet précisément. Cela me fera également travailler sur ma capacité à interagir avec des backends et des API REST, tout autant du côté front-end pour le site web que du côté SDK avec Emacs. Enfin, la création de ce SDK ainsi que des exporteurs org-mode me sera bénéfique afin d’approfondir ma connaissance d’Emacs et du Emacs Lisp. Maintenant, au travail ! ** [EN] Writing a Dynamic Array in C :dev:C: :PROPERTIES: :EXPORT_FILE_NAME: writing-dynamic-vector-c :EXPORT_DATE: 2020-11-28 :export_hugo_menu: :menu "main" :END: /Edit on October 28th 2023:/ This article was written on November 28th 2020, almost three years ago. Since then, I have noticed issues with the current implementation of my dynamic C array, as noted by some readers in the comments below. I will probably rewrite a new dynamic array in C some time in the future addressing these issues. Although C is a very, very popular language, it is also known to be quite tiny: memory is handled manually, and much of what is available in its standard library is a given in all other languages. But C being a low level language also means it lacks a lot of other stuff other popular languages have; for instance, dynamic arrays are present in the library of most popular languages, be it JavaScript, C++, Rust and so on, but C’s simplicity forbids them from being there. If you want it in C, you have to implement it –which is exactly what I did! #+TOC: headlines 1 local *** Introduction When I wrote this library, I was mostly inspired by C++’s ~std::vector~ and Rust’s ~std::vec::Vec~, but my library lacks some features both have: it’s still a simple one. Here is the list of what it is able to do: - Create a dynamic array, with or without an initial capacity specified by the user - Store a function pointer to the /destructor/ of the elements that will be stored in the vector for when they are destroyed - Append new elements at the end of the array - Get elements by position, safely or not, or get the first and last elements in the array - Get the length of the vector as well as its capacity - Shrink the size of the allocated array to the size of the vector - Remove an element at a specific index, or the last element - Completely destroy the vector and its elements Elements that will be stored in the vector will need to be dynamically allocated in memory since the vector will not store the elements themselves, but rather pointers to them. This way, we avoid copying data when inserting it to the vector, and handling these elements is also a tad easier. And since we do not know what we will be storing, we will be storing void pointers. The user will be able to cast them to their desired type later on. Before defining the vector, there are a few things I want to define. First, there is an attribute I will often use with my functions: #+NAME: vector-nonnull-h #+BEGIN_SRC c #indef NONNULL # define NONNULL __attribute__((nonnull)) #endif #+END_SRC This will forbid passing to functions marked with this attribute ~NULL~ pointers, because we will use a lot of them. We will also need to include some headers: - ~assert.h~ :: so we can make sure memory is allocated and reallocated correctly - ~string.h~ :: for some memory operations such as ~memcpy~ #+NAME: vector-includes-c #+BEGIN_SRC c #include #include #+END_SRC We also need to define a type that will be used as the destructor type. The functions we want to accept as destructors are functions that accept a void pointer to an element and return nothing, hence this definition: #+NAME: vector-destructor-type-h #+BEGIN_SRC c typedef void (*Destructor)(void *element); #+END_SRC Now, onto the structure itself. *** The Data Structure of the Vector With our vector, we will need to keep track a couple of things: - the size of the vector - the capacity of the vector - the destructor - the array itself With this, we can describe our structure for the vector: #+NAME: vector-struct-def #+BEGIN_SRC c struct Vector_s { size_t capacity; size_t length; void ** elements; Destructor destructor; }; typedef struct Vector_s Vector; #+END_SRC We have now four elements: - ~elements~ :: an array of void pointers pointing themselves either to elements stored in the vector or to nothing (initialized to ~NULL~) (note this forbids storing ~NULL~ elements in the vector), - ~length~ :: the number of elements currently stored in the vector, - ~capacity~ :: the size of the allocated memory pointed to by ~elements~ divided by the size of a void pointer. This gives us the amount of elements that can be stored in the vector without any reallocation /at most/, - ~destructor~ :: pointer to the function used to free elements stored in the vector Now, onto the functions associated with this data structure. They are all prefixed with ~vec_~ in order to avoid any collisions with other libraries and functions. *** Building Vectors The first function for building vectors is ~vec_new()~. Here is its definition: #+NAME: vector-vec_new-h #+BEGIN_SRC c Vector *vec_new(Destructor const destructor); #+END_SRC It is quite straightforward: when creating a new, standard vector, simply pass as its arguments a pointer to the destructor of this vector, either a ~NULL~ pointer for trivial data types, or a pointer to an existing function you declared somewhere. Once you do that, you get yourself a pointer to the newly created vector with which you can now store elements. Let’s see how it works under the hood: #+NAME: vector-vec_new-c #+BEGIN_SRC c Vector *vec_new(Destructor const destructor) { Vector *self; self = (Vector *)malloc(sizeof(Vector)); assert(self); ,*self = (Vector){.length = 0, .capacity = VEC_INITIAL_CAPACITY, .elements = (void *)malloc(sizeof(void *) * VEC_INITIAL_CAPACITY), .destroy = destructor}; assert(self->elements); return self; } #+END_SRC A new pointer is created, which will be the pointer returned to the user. To this pointer, we allocate enough memory to hold a vector. Once that is done, we initialize this new memory buffer with an actual vector, with its members initialized as described above. An assertion is done in order to ensure both the vector but also its storage are correctly allocated. The second function, ~vec_with_capacity~, is quite similar though not the same as ~vec_new~: it allows for an initialization of ~vec_with_capacity~ with a user-defined amount of capacity in the storage of the vector. That is, if ~vec_with_capacity(14)~ is called, the library will return a pointer to a vector which can contain and has the size of precisely fourteen elements. That way, if the user knows they’ll need a certain amount of elements to be stored in a vector, they’ll be able to reserve that exactly and limit the amount of reallocations when adding new elements. Its definition is the following: #+NAME: vector-vec_with_capacity-h #+BEGIN_SRC c Vector *vec_with_capacity(Destructor const destructor, size_t const capacity); #+END_SRC Under the hood, it calls ~vec_new~, then it will reallocate the memory already allocated for the member ~elemements~. #+NAME: vector-vec_with_capacity-c #+BEGIN_SRC c Vector *vec_with_capacity(Destructor const t_destructor, size_t const t_capacity) { Vector *self = vec_new(t_destructor); free(self->elements); (*self).elements = (void *)malloc(sizeof(void *) * t_capacity); assert(self->elements); (*self).capacity = t_capacity; return self; } #+END_SRC *** Adding Data The main feature of vectors is to hold data, so let’s make them able to take new data from the user. But first, let me explain a bit how this dynamic array which I call vector works in C. As you saw earlier, a vector is initialized with a fixed amount of memory allocated to the vector, so people can store their data in these arrays. Now, imagine you have an array of four elements, and you wish to add one more, what to do? You can reallocate your array with ~realloc~ with one more slot for your element, so now you have an array for five elements with your four original elements and a free slot for your fifth. Cool, now you can add new elements as you need them! Except that if you want to add some tens of thousands of new elements, you would end up calling some tens of thousands times ~realloc~, and that is /*slow*/. Seriously, try it, you’ll understand what I mean. And all these calls to ~realloc~ are an opportunity for it to fail. Let’s limit calls to this function, OK ? If we end up short on slots in our current array, let’s actually double the amount of slots in it. So, if we have a four-slots array, let’s make it an eight-slots array, and then a sixteen-slots array. And in a couple more calls to ~realloc~, we’ll quickly reach our tens of thousands slots array, way faster than by incrementing its capacity one by one. /“But, we’ll end up with a lot of unused memory if we need just one more element than 2^{16} elements! We don’t need a 2^{17} elements array for 2^{16}+1 elements!”/ You’re completely right, but that’s a tradeoff. Would you rather have a slow but memory-efficient program, or a fast but memory-hungry software? Plus, as you’ll see later, there is a function to shrink the size of the allocated array down to the actual amount of elements you stored in it, making it possible to temporarily have a 2^{17} elements array, and immediately after shrink it down to 2^{16}+1, once you know you won’t be adding any other elements. With this out of the way, let’s see how to add new elements to our vector. First, let’s declare a static function that reallocates the memory of a vector. Here is its declaration: #+NAME: vector-vec_realloc-def-c #+BEGIN_SRC c static void vec_realloc(Vector *const self) NONNULL; #+END_SRC Its implementation is rather simple: double its capacity, and reallocate its array twice its previous size. Of course, there is an assertion on whether the arrays have been correctly reallocated to ensure memory safety. #+NAME: vector-vec_realloc-c #+BEGIN_SRC c void vec_realloc(Vector *const self) { self->capacity *= 2; self->elements = realloc(self->elements, sizeof(void *) * vec_capacity(self)); assert(self->elements); return; } #+END_SRC Now, we can proceed to element insertion. Here is the definition of ~vec_push~, which adds a new element at the end of the vector: #+NAME: vector-vec_push-h #+BEGIN_SRC c void *vec_push(Vector *const self, void *const element) NONNULL; #+END_SRC As you can see, it takes as its arguments a pointer to the vector (the same returned by its constructor) as well as a pointer to the element to be added to the vector. This is an important point: *the vector does not store elements themselves, only their pointer*. If the function detects there is not enough space for a new element, a call will be made to ~vec_realloc~ described above. Once the function is done, it will return a pointer to the newly inserted element. #+NAME: vector-vec_push-c #+BEGIN_SRC c void *vec_push(Vector *const self, void *const t_element) { if (vec_length(self) >= vec_capacity(self)) { vec_realloc(self); } self->elements[(*self).length++] = t_element; return vec_last(self); } #+END_SRC And this is it! There may be a function added later that will allow the insertion of a new value in any valid position between the first and last position of an array (not counting the unused slots of said array), and if I implement this it will imply a reimplementation of ~vec_push~ so that ~vec_push~ relies on this potential new ~vec_insert~. *** Retrieving Data Two functions are available when retrieving data: ~vec_safe_at~ which safely retrieves the element at a certain index, and ~vec_at~, which is a bit more performant but without the safety of the former. Let’s see the definition of both: #+NAME: vector-vec_at-h #+BEGIN_SRC c void *vec_safe_at(Vector const *const self, size_t const index) NONNULL; void *vec_at(Vector const *const self, size_t const index) NONNULL; #+END_SRC Both have the same arguments: the former is a pointer to the vector we want to manipulate, and the latter is the index at which we want to retrieve our data. To see the difference in how both work, let’s first see the definition of ~vec_at~: #+NAME: vector-vec_at-c #+BEGIN_SRC c void *vec_at(Vector const *const self, size_t const index) { return self->elements[index]; } #+END_SRC ~vec_at~ is really straightforward and is just syntax sugar around the vector’s ~elements~ member and will behave exactly like the square brackets in standard C. However, ~vec_safe_at~ performs some additional checks as you can see below: #+NAME: vector-vec_safe_at-c #+BEGIN_SRC c void *vec_safe_at(Vector const *const self, size_t const t_index) { return (t_index >= vec_length(self)) ? NULL : vec_at(self, t_index); } #+END_SRC If the requested index is larger than the furthest index possible, a ~NULL~ pointer will be returned, otherwise the pointer to the requested element is. With this function, it is possible to check whether an element has been returned or not while avoiding a possible segfault or something similar. It could be used in a loop for instance in order to check we only have valid elements. It is also possible to retrieve directly the last element with ~vec_last~. Here is its definition: #+NAME: vector-vec_last-h #+BEGIN_SRC c void *vec_last(Vector const *const self) NONNULL; #+END_SRC Just as the previous functions, its declaration is really straightforward: #+NAME: vector-vec_last-c #+BEGIN_SRC c void *vec_last(Vector const *const self) { return vec_at(self, vec_length(self) - 1); } #+END_SRC For the sake of the Object-Oriented Programming paradigm, two functions were also declared in order to retrieve some data that could otherwise be easily accessible: #+NAME: vector-vec_length_capacity-h #+BEGIN_SRC c size_t vec_length(Vector const *const self) NONNULL; size_t vec_capacity(Vector const *const self) NONNULL; #+END_SRC Their implementation is extremely trivial and doesn’t really need any explanation. #+NAME: vector-vec_length_capacity-c #+BEGIN_SRC c size_t vec_length(Vector const *const self) { return self->length; } size_t vec_capacity(Vector const *const self) { return self->capacity; } #+END_SRC *** Deleting Data While this chapter is about destroying data, this first function will not exactly destroy data, or at least not data we care about: ~vec_shrink_to_fit~ will reallocate the memory in our vector to make it so that the member ~elements~ is exactly large enough to store all of our data with no more space than that. Here is its definition: #+NAME: vector-shrink_to_fit-h #+BEGIN_SRC c void vec_shrink_to_fit(Vector *const self) NONNULL; #+END_SRC There’s nothing too exciting about its implementation: a simple reallocation exactly the size of the number of elements currently stored times the size of a void pointer, and we verify with an ~assert~ if it has been correctly reallocated. Nothing is returned. #+NAME: vector-shrink_to_fit-c #+BEGIN_SRC c void vec_shrink_to_fit(Vector *const self) { if (self->length <= 0) { return; } self->capacity = self->length; self->elements = realloc(self->elements, sizeof(void *) * vec_capacity(self)); assert(self->elements); return; } #+END_SRC Notice that a check is done to see if the vector exists, because otherwise calling ~shrink_to_fit~ on an empty vector would result in an error while asserting the reallocation. Next, we have two functions: ~vec_pop_at~ and ~vec_pop~. The latter relies on the former, which can delete an element at any valid position. *Beware*: these functions return /nothing/ and simply deletes the element. Here is their definition: #+NAME: vector-vec_pop-h #+BEGIN_SRC c void vec_pop_at(Vector *const self, size_t const index) NONNULL; void vec_pop(Vector *const self) NONNULL; #+END_SRC In order to insure memory safety, a static function is declared in ~src/vector.c~ which will delete an element if a destructor has been provided to the vector when it has been built. Its definition is the following: #+NAME: vector-vec_maybe_delete_element-def-c #+BEGIN_SRC c static void vec_maybe_delete_element(Vector const *self, size_t const t_index) NONNULL; #+END_SRC Its implementation is quite simple: if a destructor exists, then the element at the requested index will be destroyed through this destructor. Otherwise, nothing is done with the destructor, hence the name of the function ~vec_maybe_delete_element~. However, it should be noted that the element will be freed from memory, so if the user needs it before popping it, they need to retrieve it with something like ~vec_at~ and store it elsewhere. #+NAME: vector-vec_maybe_delete_element-c #+BEGIN_SRC c void vec_maybe_delete_element(Vector const *self, size_t const t_index) { void *element = vec_at(self, t_index); if (self->destroy) { self->destroy(element); } free(element); } #+END_SRC Now that we have this function sorted out, we can implement our pops. Here is the implementation of ~vec_pop_at~: #+NAME: vector-vec_pop_at-c #+BEGIN_SRC c void vec_pop_at(Vector *const t_self, size_t const t_index) { if (vec_safe_at(t_self, t_index) == NULL) { return; } vec_maybe_delete_element(t_self, t_index); if (t_index + 1 < vec_length(t_self)) { memcpy(vec_at(t_self, t_index), vec_at(t_self, t_index + 1), sizeof(void *) * (t_self->length - (t_index + 1))); } --(*t_self).length; } #+END_SRC A check is performed at the beginning of the function: that the element we want to pop actually exists. If it does not, the function does nothing, otherwise the function deletes the element if needed. The call to ~vec_maybe_delete_element~ will free the requested element. Then, a check is performed to see if the requested element was at the end of the array or not. If it was not, then the elements located after the destroyed element are shifted one element closer to the beginning of the array; otherwise, if the requested element was at the end of the array, nothing is done particularly. Lastly, the count of elements stored in the vector is decreased by one. ~vec_pop~ uses the above function in order to provide a simpler call if we want to delete the last element of the array. We can see how it relies on ~vec_pop_at~ in its implementation: #+NAME: vector-vec_pop-c #+BEGIN_SRC c void vec_pop(Vector *const self) { vec_pop_at(self, vec_length(self)); } #+END_SRC Finally, ~vec_delete~ allows for the complete destruction and deallocation of a vector, including all of its elements. Here is its definition: #+NAME: vector-vec_delete-h #+BEGIN_SRC c void vec_delete(Vector *const self) NONNULL; #+END_SRC In its implementation, we can see three distinct steps: - The deletion of all its elements if a destructor exists - The deletion of the array of the vector - The deletion of the vector itself. #+NAME: vector-vec_delete-c #+BEGIN_SRC c void vec_delete(Vector *const self) { if (self->destroy) { for (size_t i = 0; i < vec_length(self); ++i) { self->destroy(self->elements[i]); } } free(self->elements); free(self); } #+END_SRC *** The Final Source Code Finally, we can see the whole source code. Here is the header for the library: ~vector.h~ #+BEGIN_SRC c :noweb yes #ifndef VECTOR_H_ #define VECTOR_H_ <> <> <> <> <> <> <> <> <> <> <> #endif /* VECTOR_H_ */ #+END_SRC And here is the implementation file: ~vector.c~ #+BEGIN_SRC c :noweb yes #include "vector.h" <> <> <> <> <> <> <> <> <> <> <> <> <> <> <> <> <> #+END_SRC And with that, we should be good! I used this library in a SOM (Kohonen, 1982) implementation and ran it through valgrind, and there were no memory leaks. If you find one though, don’t hesitate telling me in the comments, through social media such as Twitter, or by email. Happy programming! * Emacs :@emacs: ** Emacs 29 is nigh! What can we expect? :dev:emacs: :PROPERTIES: :EXPORT_FILE_NAME: emacs-29-what-can-we-expect :EXPORT_DATE: 2022-11-29 :EXPORT_OPTIONS: toc:2 :export_hugo_menu: :menu "main" :END: It [[https://lists.gnu.org/archive/html/emacs-devel/2022-11/msg01774.html][was announced a couple of hours ago]], Emacs 29’s branch is now cut from the master branch! This means the ~emacs-29~ branch will from now no longer receive any new feature, but only bug fixes. So, what’s new with this new major release? I skimmed over the ~NEWS~ file, and here are the changes which I find interesting and even exciting for some. *Article updated on December 22nd at 14:05 UTC* *** Major features A couple of major improvements will be most likely present, here are the ones that stand out the most for me. **** Eglot is now part of Emacs core During the last couple of years, LSP has given text editors incredible capabilities, giving them IDE-like features relatively easily. Aside from Elisp development, most of the code I write is now done with the help of an LSP server, running along Emacs and analysing my code, suggesting and performing changes and actions for me. Several integrations of LSP exist for Emacs, such as [[https://emacs-lsp.github.io/lsp-mode/][LSP Mode]], [[https://github.com/joaotavora/eglot][Eglot]], and [[https://github.com/manateelazycat/lsp-bridge][lsp-bridge]]. Among the three, Eglot is now part of Emacs core! No longer do you need to install a package, simply register an LSP server and autocompletion, documentation, error detection, and other features will become available right away! I must admit I don’t really know Eglot, I personally use LSP Mode, but with this addition to Emacs core, I might attempt the switch. **** Tree-Sitter is also part of Emacs core In case you didn’t know, Emacs’ current syntax highlighting is currently based on a system of regexes. Although it is not the /worst/ thing to use, it’s not the best either, and it can become quite slow on larger files. Tree-Sitter parses programming languages based into a concrete syntax tree. From there, not only can syntax highlighting can be done at high speed, but a much deeper analysis of the code is possible and actions such as syntax manipulation can also be achieved since the syntax tree itself is available as an object which can be manipulated! In case you want some more information on Tree-Sitter itself, you can check out the [[https://tree-sitter.github.io/tree-sitter/][official Tree-Sitter website]], or you can even check this talk out given by Tree-Sitter’s creator, Max Brunsfeld. #+begin_export html #+end_export Well, this is now a native solution in Emacs! Currently, Emacs’ Tree-Sitter supports the current major modes : - ~bash-ts-mode~ - ~c-ts-mode~ - ~c++-ts-mode~ - ~csharp-ts-mode~ - ~css-ts-mode~ - ~java-ts-mode~ - ~js-ts-mode~ - ~json-ts-mode~ - ~python-ts-mode~ - ~typescript-ts-mode~ Tree-Sitter also holds for now a special status in the new ~emacs-29~ branch since new features can still be added to it, as its merge with the master branch is still recent. So we might see the list of major modes for Emacs get a bit longer yet, especially considering Tree-Sitter tries to make adding new languages relatively easy. If you can’t wait to test Tree-Sitter, there is already [[https://emacs-tree-sitter.github.io/][another package]] available for Emacs you can use right now. Just be aware this is not the same package as the one that got integrated into Emacs. **** Install packages from source with ~package.el~ If you use [[https://github.com/radian-software/straight.el][Straight]], you might be familiar with installing packages directly from their Git repository. Well, good news, it is now possible to install packages from Git using Emacs’ built-in packaging system ~package.el~! It can be done with the new function ~package-vc-install~, and packages installed that way can be updated with ~package-vc-update~ or ~package-vc-update-all~. On the topic of ~package.el~, there is also the new function ~package-report-bug~ which allows Emacs users to report bugs to the developers of a package directly from Emacs! Be aware though, it only works for packages installed through ~package.el~. Since I’m a [[https://github.com/jwiegley/use-package][~use-package~]] and ~straight.el~ user, there is no package listed when I invoke the command. **** Org mode 9.6 As confirmed by one of org-mode maintainers [[https://bzg.fr/][Bastien Guerry]] on [[https://lists.sr.ht/~bzg/emacsfr/%3C87bkophdzo.fsf%40phundrak.com%3E#%3C87tu2hqf4h.fsf@gnu.org%3E][a French-speaking Emacs mailing list]], Org 9.6 is set to be part of Emacs 29! There is an [[https://orgmode.org/Changes.html][official article]] on this release, which is [[https://elpa.gnu.org/packages/org.html][already available on GNU ELPA]]! **** use-package in Emacs core It has also been confirmed on the [[https://lists.gnu.org/archive/html/emacs-devel/2022-11/msg01821.html][Emacs development mailing list]] that [[https://github.com/jwiegley/use-package][~use-package~]], an awesome package manager, is set to be part of Emacs 29, although it initially wasn’t included in the ~emacs-29~ branch. **** Pure GTK Emacs is here for Wayland! One of the major issues Emacs had on Linux was its dependency on Xorg when running in GUI mode. When running Xorg, it’s not really an issue, but Wayland has become more and more common during the last years, and even with the existence of XWayland, this became an annoyance. Well, fear not, for pure GTK Emacs is here! It can now be built Xorg-free and run natively in Wayland! Be aware though that Wayland is basically the only use-case for pure GTK Emacs. If you don’t use Wayland, Emacs will display a warning message, as it will most likely cause issues if you are running Xorg. In my case, I sometimes see some ghost text when the content of a buffer updates (I still need pure GTK though, since I alternate between Xorg and Wayland). **** Compile EmacsLisp files ahead of time With Emacs 28 came the ability to natively compile EmacsLisp if your Emacs was built with the ability to do so, using GCC’s Just In Time library. This results in quite the impressive boost in performance, which made Emacs much snappier than it was before. The only issue I had was Emacs would only compile its EmacsLisp files when they were loaded for the first time. This is no longer the case! If you now compile Emacs with ~--with-native-compilation=aot~, Emacs’ native EmacsLisp files will be natively compiled along with Emacs itself! Be aware though, it can be slow on most machines, so the time you save by not compiling these files when launching Emacs for the first time is basically transferred to when compiling Emacs itself. Is it worth your time? In my case, I would say yes, because when I compile Emacs, I’m generally not in a hurry. But in your case? Well, test it out and see for yourself. **** Native access to SQLite databases Emacs can now be built with native support for SQLite and the sqlite3 library. In fact, this is now a default behaviour, since you need to pass ~--without-sqlite3~ to Emacs’ build configuration script in order to prevent it. This comes with a new ~sqlite-mode~ which allows you to explore SQLite databases within Emacs and to interact with them. Check out the ~sqlite-mode-open-file~ function! **** HaikuOS support For all three HaikuOS users out there, good news, you now have access to Emacs! (In all seriousness, I should check out HaikuOS one day) Moreover, it also supports an optional window-system port to Haiku with ~--with-be-app~. Be aware, you will need the Haiku Application Kit development headers and a C++ compiler. Otherwise, Emacs will only run in the terminal. If you want to also add Cairo to the mix, you can add ~--with-be-cairo~. **** New major mode for C# ~csharp-mode~ is now a native major mode for Emacs and is based on ~cc-mode~. *** Minor features **** It’s easier to use Emacs in scripts! If you like to write scripts and especially writing Lisp scripts, Emacs now supports the option ~-x~ in order to execute scripts written in EmacsLisp. When executing such a script with ~#!/usr/bin/emacs -x~ as its shebang, Emacs will not read its init file (like with ~-Q~) and will instead execute the Elisp code right away and return the last value to the caller of the script (most likely the shell you called the script from). **** TRAMP natively supports Docker, Podman, and Kubernetes Three new connections are now available for TRAMP: - ~docker~ - ~podman~ - ~kubernetes~ You will now be able to access your containerized environment right from Emacs without the need to write custom code. **** Custom user directory It is now easier to launch custom Emacs profiles without the need of tools such as [[https://github.com/plexus/chemacs2][chemacs2]] with the addition of the flag ~--init-directory~. This can set to any directory Emacs’ ~user-emacs-directory~ which includes the ~init.el~ which comes along with it. Yet another reason for me not to use a ~.emacs~ file, but the ~init.el~ file instead. **** Support for Webp images For quite some time, Emacs has been able to display images, but not webp yet. Well, this is now fixed! And in fact, support for webp images became the default behaviour, since you need to pass ~--without-webp~ to Emacs’ configuration script to disable webp support. **** C++ mode now supports the C++20 standard Yep. There’s nothing more to say, really. Happy coding! **** Better handling of ~.pdmp~ files Emacs has had for a few version the ability to dump its state into a ~pdmp~ file for faster startup time. Well now, when creating such a file, it will include in its name a fingerprint of its current state, although it will still prioritize an ~emacs.pdmp~ file if it exists. **** Better mouse and touchpad support Emacs now uses XInput 2, which enables Emacs to support more input events, such as touchpad events. For instance, by default, a pinch gesture on a touchpad increases or decreases the text size of the current buffer. This is thanks to the new event ~pinch~, which comes along with ~touch-end~. **** Unicode 15.0 and emojis Emacs now supports [[https://www.unicode.org/versions/Unicode15.0.0/][Unicode 15.0]], which is currently the latest Unicode version. Although this is not directly related, quite a few new emoji-related features have been introduced. The new prefix ~C-x 8 e~ now leads to a few new commands related to emojis: - ~C-x 8 e e~ or ~C-x 8 e i~ :: Insert an emoji (~emoji-insert~) - ~C-x 8 e s~ :: Search an emoji (~emoji-search~) - ~C-x 8 e l~ :: List all emojis in a new buffer (~emoji-list~) - ~C-x 8 e r~ :: Insert a recently inserted emoji (~emoji-recent~) - ~C-x 8 e d~ :: Describe an emoji (~emoji-describe~) - ~C-x 8 e +~ and ~C-x 8 e -~ :: Increase and decrease the size of any character, but especially emojis (~emoji-zoom-increase~ and ~emoji-zoom-decrease~ respectively) There is also the new input method ~emoji~ which allows you to type for instance ~:⁣grin:~ in order to get the emoji 😁. **** True background transparency Up until recently, if you wanted transparency with Emacs, you had no choice but to make the whole frame transparent, including text and images. Thanks to the frame parameter ~alpha-background~ and its related ~alphaBackground~ X resource, it is now possible to set transparency only for the frame’s background without affecting any of the other elements on screen. **** WebKit inspector in Emacs’ WebKit widget browser You can now access the WebKit inspector when using the WebKit widget browser in Emacs, given you are using a version of Emacs which has been compiled with it. I wish there was a keybinding or at least a function for it, but apparently you can only open it with a right click and select /Inspect Element/. Still nice to have. **** Some news for Windows Although it has been available for Linux users since Emacs 26.1, Windows finally has access to double-buffering to reduce display flicker. If you wish to disable it, you can set the frame parameter ~inhibit-double-buffering~ to ~nil~. Emacs also follows Windows’ dark mode with Windows 10 (version 1809) and onwards. Emacs also now uses Windows’ native API to render images. This includes BMP, GIF, JPEG, PNG, and TIFF images. Other formats, however, still rely on other dependencies and libraries to properly work, such as Webp images. *** What’s next? With Emacs 29 being cut, development on the master branch will now go towards Emacs 30. Is there anything we can expect yet? It’s still very early to say, most stable features merged into master went to Emacs 29, and only the ~feature/pkg~ and ~feature/improved-lock-narrowing~ branches seem to have received commits less than a week prior to the day of writing this, and I do not know the status of other branches that received commits during the past few weeks such as ~feature/package+vc~ or ~feature/eglot2emacs~ (which I assume both got merged). +However, there are currently talks about including ~use-package~ into Emacs! I’m a bit disappointed it won’t make it into Emacs 29, but progress is being made on ~scratch/use-package~, and you can always check the mailing list to check its status such as [[https://lists.gnu.org/archive/html/emacs-devel/2022-11/msg01533.html][here]].+ *Update*: Rejoice! As mentioned above, ~use-package~ is actually set to land in Emacs 29! ** [EN] Automatic Meaningful Custom IDs for Org Headings :emacs:orgmode:dev: :PROPERTIES: :EXPORT_FILE_NAME: better-custom-ids-orgmode :EXPORT_DATE: 2020-06-06 :export_hugo_menu: :menu "main" :END: Spoiler alert, I will just modify a bit of code that already exists, go directly to the bottom if you want the solution, or read the whole post if you are interested in how I got there. #+TOC: headlines 1 local **** Update 2021-11-22 I’ve put the code presented here as a complete package. You can find it in [[https://labs.phundrak.com/phundrak/org-unique-id][this repository]] or in its [[https://github.com/Phundrak/org-unique-id][GitHub mirror]] (be aware the latter may not be as up-to-date as the former is. Installation instructions are in the README. *** The issue About two to three years ago, as I was working on a project that was meant to be published on the internet, I looked for a solution to get fixed anchor links to my various headings when I performed HTML exports. As some of you may know, by default when an Org file is exported to an HTML file, a random ID will be generated for each header, and this ID will be used as their anchor. Here’s a quick example of a simple org file: #+caption: Example org file #+begin_src org :exports code ,#+title: Sample org file ,* First heading Reference to a subheading ,* Second heading Some stuff written here ,** First subheading Some stuff ,** Second subheading Some other stuff #+end_src And this is the result once exported to HTML (with a lot of noise removed from ~~): #+caption: Output HTML file #+BEGIN_SRC html Sample org file

Sample org file

1 First heading

Reference to a subheading

2 Second heading

Some stuff written here

2.1 First subheading

Some stuff

2.2 Second subheading

Some other stuff

#+END_SRC As you can see, all the anchors are in the format of ~org[a-f0-9]{7}~. First, this is not really meaningful if you want to read the anchor and guess where it will lead you. But secondly, these anchors will change each time you export your Org file to HTML. If I want to share a URL to my website and to a specific heading, … well I can’t, it will change the next time I update the document. And I don’t want to have to set a ~CUSTOM_ID~ property for each one of my headings manually. So, what to do? *** A first solution A first solution I found came from [[https://writequit.org/articles/emacs-org-mode-generate-ids.html][this blog post]], where Lee Hinman described the very same issue they had and wrote some Elisp code to remedy that (it’s a great read, go take a look). And it worked, and for some time I used their code in my Emacs configuration file in order to generate unique custom IDs for my Org headers. Basically what the code does is it detects if ~auto-id:t~ is set in an ~#+OPTIONS~ header. If it is, then it will iterate over all the Org headers, and for each one of them it will insert a ~CUSTOM_ID~, which is made from a UUID generated by Emacs. And tadah! we get for each header a ~h-[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}~ custom ID that won’t change next time we export our Org file to HTML when we save our file, and only for headings which don’t already have a ~CUSTOM_ID~ property. Wohoo! Except… *** These headers are not meaningful OK, alright, that’s still a huge step forward, we don’t have to type any ~CUSTOM_ID~ property manually any more, it’s done automatically for us. But, when I send someone a link like ~https://langue.phundrak.com/eittland#h-76fc0b91-e41c-42ad-8652-bba029632333~, the first reaction to this URL is often something along the lines of “What the fuck?”. And they’re right, this URL is unreadable when it comes to the anchor. How am I supposed to guess it links to the description of the vowels of the Eittlandic language? (That’s a constructed language I’m working on, you won’t find anything about it outside my website. Also, this link is dead now, it got simplified thanks to Vuepres.) So, I went back to my configuration file for Emacs, and through some trial and error, I finally found a way to get a consistent custom ID which is readable and automatically set. With the current state of my code, what you get is the complete path of the Org heading, all spaces replaced by underscores and headings separated by dashes, with a final unique identifier taken from an Emacs-generated UUID. Now, the same link as above will look like ~https://langue.phundrak.com/eittland#Aperçu_structurel-Inventaire_phonétique_et_orthographe-Voyelles_pures-84f05c2c~. It won’t be more readable to you if you don’t speak French, but you can guess it is way better than what we had before. I even added a safety net by replacing all forward slashes with dashes. The last ID is here to ensure the path will be unique in case we’d have two identical paths in the org file for one reason or another. The modifications I made to the first function ~eos/org-id-new~ are minimal, where I just split the UUID and get its first part. This is basically a way to simplify it. #+BEGIN_SRC emacs-lisp (defun eos/org-id-new (&optional prefix) "Create a new globally unique ID. An ID consists of two parts separated by a colon: - a prefix - a unique part that will be created according to `org-id-method'. PREFIX can specify the prefix, the default is given by the variable `org-id-prefix'. However, if PREFIX is the symbol `none', don't use any prefix even if `org-id-prefix' specifies one. So a typical ID could look like \"Org-4nd91V40HI\"." (let* ((prefix (if (eq prefix 'none) "" (concat (or prefix org-id-prefix) "-"))) unique) (if (equal prefix "-") (setq prefix "")) (cond ((memq org-id-method '(uuidgen uuid)) (setq unique (org-trim (shell-command-to-string org-id-uuid-program))) (unless (org-uuidgen-p unique) (setq unique (org-id-uuid)))) ((eq org-id-method 'org) (let* ((etime (org-reverse-string (org-id-time-to-b36))) (postfix (if org-id-include-domain (progn (require 'message) (concat "@" (message-make-fqdn)))))) (setq unique (concat etime postfix)))) (t (error "Invalid `org-id-method'"))) (concat prefix (car (split-string unique "-"))))) #+END_SRC Next, we have here the actual generation of the custom ID. As you can see, the ~let~ has been replaced by a ~let*~ which allowed me to create the ID with the variables ~orgpath~ and ~heading~. The former concatenates the path to the heading joined by dashes, and ~heading~ concatenates ~orgpath~ to the name of the current heading joined by a dash if ~orgpath~ is not empty. It will then create a slug out of the result, deleting some elements such as forward slashes or tildes, and all whitespace is replaced by underscores. It then passes ~heading~ as an argument to the function described above to which the unique ID will be concatenated. #+BEGIN_SRC emacs-lisp (defun eos/org-custom-id-get (&optional pom create prefix) "Get the CUSTOM_ID property of the entry at point-or-marker POM. If POM is nil, refer to the entry at point. If the entry does not have an CUSTOM_ID, the function returns nil. However, when CREATE is non nil, create a CUSTOM_ID if none is present already. PREFIX will be passed through to `eos/org-id-new'. In any case, the CUSTOM_ID of the entry is returned." (interactive) (org-with-point-at pom (let* ((orgpath (mapconcat #'identity (org-get-outline-path) "-")) (heading (replace-regexp-in-string "/\\|~\\|\\[\\|\\]" "" (replace-regexp-in-string "[[:space:]]+" "_" (if (string= orgpath "") (org-get-heading t t t t) (concat orgpath "-" (org-get-heading t t t t)))))) (id (org-entry-get nil "CUSTOM_ID"))) (cond ((and id (stringp id) (string-match "\\S-" id)) id) (create (setq id (eos/org-id-new (concat prefix heading))) (org-entry-put pom "CUSTOM_ID" id) (org-id-add-location id (buffer-file-name (buffer-base-buffer))) id))))) #+END_SRC The rest of the code is unchanged, here it is anyway: #+BEGIN_SRC emacs-lisp (defun eos/org-add-ids-to-headlines-in-file () "Add CUSTOM_ID properties to all headlines in the current file which do not already have one. Only adds ids if the `auto-id' option is set to `t' in the file somewhere. ie, #+OPTIONS: auto-id:t" (interactive) (save-excursion (widen) (goto-char (point-min)) (when (re-search-forward "^#\\+OPTIONS:.*auto-id:t" (point-max) t) (org-map-entries (lambda () (eos/org-custom-id-get (point) 'create)))))) (add-hook 'org-mode-hook (lambda () (add-hook 'before-save-hook (lambda () (when (and (eq major-mode 'org-mode) (eq buffer-read-only nil)) (eos/org-add-ids-to-headlines-in-file)))))) #+END_SRC Note that you *will need* the package ~org-id~ to make this code work. You simply need to add the following code before the code I shared above: #+BEGIN_SRC emacs-lisp (require 'org-id) (setq org-id-link-to-org-use-id 'create-if-interactive-and-no-custom-id) #+END_SRC And that’s how my links are now way more readable *and* persistent! The only downside I found to this is when you move headings and their path is modified, or when you modify the heading itself, the custom ID is not automatically updated. I could fix that by regenerating the custom ID on each save, regardless of whether a custom ID already exists or not, but it’s at the risk an ID manually set will get overwritten. * Linux :@linux: ** [EN] My YouTube subscriptions as an RSS feed :linux:dev:tutorial: :PROPERTIES: :EXPORT_FILE_NAME: youtube-subscriptions-rss :EXPORT_DATE: 2022-02-04 :export_hugo_menu: :menu "main" :END: *** The Problem I’m sure you’ve been in the same situation before: you go on YouTube because you want to watch a video, maybe two, from your subscriptions. You open the first one. Oh great, an unskippable fifteen seconds ad. And another one! OK, the video starts. It gets cut a couple of times by other ads of varying length. Oh but what’s this? This recommended video looks nice! And before you know it, your whole afternoon and evening went by painfully watching videos on YouTube’s atrocious video player. You lost focus. *** My Solution: mpv + RSS Wouldn’t it be nice if it were possible to watch these videos with a full-fledged video player over which you have complete control? Which could be customized to your heart’s content? Which won’t secretly track what you watch? Oh right, [[https://mpv.io/][mpv]]! It supports most video formats you can think of, and thanks to its interoperability with [[https://github.com/ytdl-org/youtube-dl][youtube-dl]], you can also watch videos from [[https://ytdl-org.github.io/youtube-dl/supportedsites.html][an extremely wide variety of websites]]! So why not YouTube? Now, the question is how to get rid of YouTube’s interface. The answer is actually quite simple: let’s use an RSS feed. With the RSS feeds from YouTube, you will receive in your RSS reader the link of the video with its thumbnail and its description. You can then copy from there the link and open it with mpv with a command like this: #+begin_src bash mpv "https://www.youtube.com/watch?v=xym2R6_Qd7c" #+end_src **** Channel RSS Now the question is how to get the RSS feed of a channel? The answer is quite simple. The base URL for a YouTube channel RSS feed is ~https://www.youtube.com/feeds/videos.xml?channel_id=~ to which you simply have to add the channel ID. For instance, if you want to follow Tom Scott with this, you simply have to extract the part of the channel after ~/channel/~ in his URL and append it to the URL mentioned above, and TADAH! you get an RSS feed to his channel! #+begin_src text https://www.youtube.com/feeds/videos.xml?channel_id=UCBa659QWEk1AI4Tg--mrJ2A #+end_src Be careful to select the channel ID only if it is after a ~/channel/~ though! The part that is after a ~/c/~ will not work. If you end up on the URL ~https://www.youtube.com/c/TomScottGo~, simply click on a random video, then click on the channel’s name. This should bring you back to the channel but with an important difference: the URL is now ~https://www.youtube.com/channel/UCBa659QWEk1AI4Tg--mrJ2A~. The thing that is really nice with this setup is you don’t really need to actually subscribe to a channel, your RSS feed already does that for you! And with lots of RSS feed readers, you can categorize your different feeds, meaning you can even categorize your subscriptions! **** Playlist RSS It is also possible to follow not only a channel but a playlist of videos. For that, you will instead use ~https://www.youtube.com/feeds/videos.xml?playlist_id=~ as your base URL to which you will add the ID of the playlist you want to follow. For instance, with Tom Scott’s playlist for Citation Needed Season 7, the URL of the playlist is ~https://www.youtube.com/playlist?list=PL96C35uN7xGI15-QbtUD-wJ5-G8oBI-tG~, which means you need to keep the ~PL96C35uN7xGI15-QbtUD-wJ5-G8oBI-tG~ and put it into the URL like so: #+begin_src text https://www.youtube.com/feeds/videos.xml?playlist_id=PL96C35uN7xGI15-QbtUD-wJ5-G8oBI-tG #+end_src *** Which RSS reader to go with? If you know me, you’ll know I am extremely biased towards Emacs, so of course I’ll recommend Elfeed to any Emacs user ([[https://config.phundrak.com/emacs#Packages-Configuration-Applications-Elfeedoip0fl6184j0][my relevant configuration is here]]). I even wrote an advice around ~elfeed-show-visit~ to ensure YouTube videos are open with mpv instead of my web browser. If you’re not into Emacs, or not /that/ into Emacs, you can also try other alternatives such as [[https://gitlab.com/news-flash/news_flash_gtk][NewsFlash]], a very nice RSS reader written in GTK for Linux –I may not always agree with DistroTube, but he made a [[https://www.youtube.com/watch?v=KBAmviddh4A][very nice video]] presenting this piece of software. (Remember, right-click and then ~mpv "the url here"~!) The [[https://apps.nextcloud.com/apps/news][News app]] for Nextcloud is also very neat, I recommend you using it. You can also get your RSS feed in your terminal with [[https://newsboat.org/][Newsboat]]. Not really my cup of tea, but I can see why some people enjoy it. *** Improving a bit the mpv tooling You might have heard it, but youtube-dl hasn’t been doing great recently. The tool is becoming slow, and it lacks quite a few features it could really benefit from. While it is important to acknowledge its historical importance, I think it is now time to move on, and its successor shall be [[https://github.com/yt-dlp/yt-dlp][yt-dlp]]. In my experience, this youtube-dl fork is much faster than youtube-dl itself on top of providing additional features such as [[https://github.com/yt-dlp/yt-dlp#sponsorblock-options][SponsorBlock integration]]. How do you replace youtube-dl with yt-dlp then? If you use Arch Linux or one of its derivates (I hope not Manjaro though), you can simply install ~yt-dlp-drop-in~ from the AUR. #+begin_src bash paru -S yt-dlp-drop-in # or if you prefer yay yay -S yt-dlp-drop-in # or whichever AUR helper you prefer, as long as it is NOT yaourt #+end_src If you are not an Arch Linux user, check out [[https://www.funkyspacemonkey.com/replace-youtube-dl-with-yt-dlp-how-to-make-mpv-work-with-yt-dlp][this article]], it will help you.