Movatterモバイル変換


[0]ホーム

URL:


Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

Sign up
Appearance settings

Allow hooking into the filesystem and providing a persistent backend (like absurd-sql)#481

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to ourterms of service andprivacy statement. We’ll occasionally send you account related emails.

Already on GitHub?Sign in to your account

Open
jlongster wants to merge5 commits intosql-js:master
base:master
Choose a base branch
Loading
fromjlongster:master

Conversation

@jlongster
Copy link

This is (finally) a follow-up to myabsurd-sql project I talked about here:#447 (comment)

These are the changes so far required to make it work. It's not a lot of changes, but there are several small things:

  • We need to stabilize the FS exports to projects can hook into the filesystem. Long-term, I'd like to figure out a better way to do this. Emscripten has plans to change their FS implementation a lot, so it will likely change. For now, this lets us hook in the FS that's already there. We do that by specifying some exports so closure doesn't minify them.
  • We add afilename to theDatabase constructor which is a boolean specifying that you've passed in a filename, not a buffer. Open to a better API here.
  • Don't delete the DB on close (if it's not a buffer), haha. If Obviously now that it persists we don't want that.
  • Perhaps the ugliest part, we add a little bit of C code to customize the default unix vfs. One reason to do this is to customizeblockDeviceCharacteristics which might improve performance characteristics. The bigger reason is to hook into the lock/unlock methods of the vfs and override them. Unfortunately, this requires some ugly code inapi.js to map the internal sqlite file to a filename and then call into our FS to handle them.

Normally, we could handle thefcntl system call at the FS layer. Unfortunately, right now emscripten doesn't provide a way to do that. If it did, we wouldn't need the last point. We might still want to overrideblockDeviceCharacteristics, but that could be an optional stop to improve performance. Right now it's critical so we get locking.

I've filed an issue with emscripten:emscripten-core/emscripten#15070. This PR could be simplified once we have that. Not sure if you want to wait on that to land this PR or now.

Anyway, I'm open to how you want to proceed. While absurd-sql is robust and ready for production (I'm using it in prod), the API and implementation here is somewhat proof of concept. Let me know what you think.

alex-kinokon, quolpr, Taytay, elazar, steida, dpmccabe, webness, monsendag, adrianoconnor, crstffr, and 25 more reacted with thumbs up emoji0xcaff, Miaourt, alex-kinokon, Taytay, steida, typfel, 0xtsukino, frankleng, Schmavery, pboling, and pedrox-hs reacted with hooray emojiMiaourt, brynne8, steida, smithbm2316, monsendag, typfel, 0xtsukino, Circuit8, frankleng, ngshiheng, and 2 more reacted with rocket emoji
@Taytay
Copy link
Contributor

Nice@jlongster

What do you think@lovasoa? I'm excited about this PR on principle.

@lovasoa
Copy link
Member

Hello everyone!
I've been wanting to review this PR for a long time, but never found the time to get to it.
I realize that sql.js is stagnating mainly because of me, and I would like to add a new maintainer.@jlongster, you have been doing really awesome work around this project. Do you think you would have the time and energy to be a maintainer?

elazar, steida, JonathanMontane, webness, adrianoconnor, crstffr, rogerfar, stefnotch, frankleng, biels, and 5 more reacted with thumbs up emoji

@arsnyder16
Copy link

@lovasoa Any luck getting a new maintainer or a chance to review this? Would be a great to get these changes in so absurd sql could be used without using a forked version of sql.js

DannyArmstrong, steida, richiksc, bkotsopoulossc, Taytay, abhishiv, joekukish, fuintis, tianjianchn, stefnotch, and 3 more reacted with heart emoji

@quolpr
Copy link

@lovasoa could you please review it? This change will make a big move in using of sql.js for the web 🙏

delaneyj, smithbm2316, shibasis0801, quentez, and Faolain reacted with thumbs up emoji

@tantaman
Copy link

Hey@lovasoa - is there any way I can help here?

I can do a cursory review but I'm not deeply familiar with the internals ofsql.js yet -- I hope to be there eventually.

I'm interested in the future ofsql.js as the developer and maintainer ofhttps://aphrodite.sh/ (multi-platform / local-first ORM)

delaneyj, stefnotch, smithbm2316, shibasis0801, and quentez reacted with thumbs up emoji

@jimmywarting
Copy link

considering that we now havehttps://github.com/whatwg/fs then i think indexeddb is rather 👎

@mfulton26
Copy link

@jimmywarting any idea on whenStorageManager.getDirectory() will become available on Chrome for Android?

This sounds extremely promising but I need mobile support. I couldn't find any mention of this inChrome Platform Status orIssues - chromium.

StorageManager API: getDirectory | Can I use... Support tables for HTML5, CSS3, etc

@mfulton26
Copy link

@jimmywarting
Copy link

i don't know.
technically there isn't any mobile specific UI issues about picking files etc, whatwg fs is purely about the api and adding a sandboxed fs
The "File system access" on the other hand is for a mean to request access to a user selected file or folder and requires UI for prompting access. The android issue are here:https://crbug.com/1011535

safari is actively working on whatwg fs behind a expe flag, but it sucks pretty much right now. You can only create folders and empty files. you can't read or write data to/from files atm.

i haven't heard anything from FF yet.

Anyway. i just learned thatnavigator.storage.getDirectory() is just an alternative new api forwebkitRequestFileSystem(TEMPORARY, 0, success, fail)
modifying files in webkitRequestFileSystem will make an effect onnavigator.storage.getDirectory and also the other way around

the benefit of usingnavigator.storage.getDirectory in web worker are that you can have both a async and sync variant

@jimmywarting
Copy link

@jimmywarting
Copy link

also have a look athttps://github.com/WICG/file-system-access/issues

mfulton26 reacted with thumbs up emoji

@mfulton26
Copy link

I considered closing it as a duplicate but then I realized that as the Origin Private File System requires no user interactions nor permission granting that it might be best to track it separately. I had not found that other crbug though. Thank you for all the links!

@rhashimoto
Copy link

I've built SQLite filesystems with both Origin Private File System access handles and IndexedDB. In my benchmarks OPFS was somewhat slower. This may in part be because I used non-synchronous access handles for reading database files (and synchronous handles for writing) to allow concurrent read access from multiple database connections, but is also because atomic commit and optional durability settings provide some implementation advantages for IndexedDB.

That said, unless performance or browser support is the overriding priority, I would much rather write code for OPFS than IndexedDB.

jljorgenson18, quentez, and MegaBlackLabel reacted with thumbs up emoji

@mrmurphy
Copy link

The file system standard looks like, but as far as I can tell from reading around the web, Firefox has no plans to implement support. For the moment, anyway, IDB seems to be the most compatible backend for SQL.js. Is this PR held up on review because of the file system access standard?

milahu reacted with thumbs up emoji

@rhashimoto
Copy link

Firefox has objections to the full File System Access API proposal, but has beenmore encouraging on the parts used for Origin Private File System, including the AccessHandle API (which is what Safari has implemented). Theirimplementation activity has stepped up in the last few weeks so it looks like they are actively working on at least trial support.

jimmywarting and milahu reacted with thumbs up emoji

@mrmurphy
Copy link

Well that's very encouraging!

@SergNikitinSergNikitin mentioned this pull requestDec 8, 2022
8 tasks
@biels
Copy link

Any news on this?https://caniuse.com/?search=opfs Apparently OPFS is avaliable on most browsers now and one intended use case is actually as storage for sqlite-wasm databases in web workers

@rhashimoto
Copy link

Both IndexedDB and OPFS access handles can be used for persistent SQLite storage in the browser. At the moment it appears that neither is clearly better than the other for this purpose; it depends on your application requirements.

  • IndexedDB
    • Lower write transaction overhead.
    • Allows concurrent reads.
    • Allows trading durability for performance.
  • OPFS access handles
    • Faster raw reads and writes.
    • Allows synchronous VFS implementation (i.e. without Asyncify or SharedArrayBuffer).

@jimmywarting
Copy link

jimmywarting commentedApr 26, 2023
edited
Loading

or SharedArrayBuffer

Oh, that reminds me. I just recently wanted to port something NodeJS-ish that used syncnode:crypto stuff to handle things to make it more cross env. friendly for Deno and web. so i wanted to switch it out to web crypto instead now that NodeJS has it too.

But it was to annoying to switch the entire codebase to be async-ified and breaking all the existing tests so i builtawait-sync to make the process a bit easier to use.

i think you@rhashimoto would maybe find this interesting and perhaps useful to transform async code to sync more easily.
...It usesSharedArrayBuffer behind the scenes. so it only works in backends and also web worker.

(a bit off topic, but i wanted to share it...)

@krisgrm
Copy link

Following this one

@clicktodev
Copy link

is it currently possible to save a sql.js db to opfs?

Sign up for freeto join this conversation on GitHub. Already have an account?Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

13 participants

@jlongster@Taytay@lovasoa@arsnyder16@quolpr@tantaman@jimmywarting@mfulton26@rhashimoto@mrmurphy@biels@krisgrm@clicktodev

[8]ページ先頭

©2009-2025 Movatter.jp