Ability to resume failed/interrupted downloads in progress

[Fironet]

While attempting to download a soundtrack for a release on itch.io unsuccessfully (as part way through it fails) I've found itch.io doesn't allow downloads to be resumed. For large downloads especially it can be frustrating. In this case it was a 327MB FLAC download on Firefox v46/Win 8.1.

I frequently use Bandcamp, and for comparison it allows for resuming downloads when similar download failures occur, something which is highly appreciated for large downloads.

Adding such an ability would be a time-saving and worthwhile improvement to the download functionality of the site and help the user from having to re-download the file again.

[fasterthanlime]

@Fironet note that https://itch.io/app allows resuming downloads.

The reason a browser won't be able to resume an itch.io download, for example, is that the download URL is a signed URL that expires after a short time.

That said, I think we could have downloads work via a redirect (HTTP 307 seems fitting) — when downloading from a browser, with the user logged in, we can easily re-generate a signed download URL for them to resume the download.

The only question is whether browser will request the first URL (permanent) or the second URL (temporary/signed/expiring) when trying to resume the download. If they try the latter, this solution isn't going to work.

[Fironet]

@Fironet note that https://itch.io/app allows resuming downloads.

Tried this a few days ago. Probably was made to be seamless for most users, though I wasn't fond of how it auto installed without a prompt rather than asking where I'd like to install it.

Any method to improve downloads for the web interface would be appreciated 👍 Ended up completing the download mentioned above in Opera 12 of all things. Seems to be more reliable with such downloads, or perhaps it just didn't experience any network interruptions.

[fasterthanlime]

auto installed without a prompt rather than asking where I'd like to install it.

Exactly :) power users can manage install locations in Preferences.

[Efreak]

I'd like to note that until downloads are resumable in the browser, those of us with poor network connections cannot download larger games on Android without a data connection. You can't download apk files in the desktop client, and you can't use a download manager. Therefore, the only options are to spend hours repeatedly, manually downloading the same file every time it gets stuck at 10%, or to not get a game. Please, please make resumable downloads work. Better yet would be an itch.io client for Android.

[ultimaterex]

I'm surprised that an Issue this old has not been addressed, but please allow users to download from the desktop itch client. Currently the flow only offers installing the platform specific version, this would at least offer android users a way to resume downloads.

[Haerbin23456]

I even need to retry 2 times before successfully download the desktop itch client itself. wtf

[dgatwood]

This is a rather serious bug that seems to be caused by the way they use AWS. Instead of having a permanent URL, the URL uses an authorization code that has an expiration date just a couple of minutes into the future. When the client goes to resume the download, the expiration has already passed, and it is impossible to resume the download without going through the JavaScript code again to create a new URL, which the browser obviously cannot do.

Compounding matters is the fact that a lot of modern browsers are much more aggressive at canceling stalled HTTP requests for downloads under the assumption that it should be resumable. So on slow networks, this can make it nearly impossible to download from Itch.io unless you very quickly start a download in the browser, copy the URL, cancel the download, and retry it using curl within the narrow timeout window.

To fix this:

• Set the expiration date an hour into the future instead of a minute or two.
• Ensure that the web server properly supports byte ranges and the ETag header.

Chances are the second one "just works", which means this can probably be fixed by just making a single constant larger somewhere in the code base.