Make Mojo::Asset::File writable to the existed files.

[iakuf]

After debugging for days, I figured it out finally and fixed my code.
This patch is to make the Mojo::Asset::File more flexible, it gives the write permission to the module when handling uploading/downloading files.

[kraih]

What's the use case for this change?

[iakuf]

Actually, when uploading a huge file(eg. 2GB), I hope Mojo::Asset::File to be able to upload the file in multiply HTTP requests. In this way, people won't have to wait the uploading process for hours, instead, they can resume the upload process at any time they want.

I found the "add_chunk" in Mojo::Asset::File is able to do it. Unfortunately, "add_chunk" can only add "chunk" to the file it's uploading. It's not able to add chunk to an existed file, which is required for my use case.

So my proposal is to only change one bit of Mojo::Asset::File module at line 20. It will help a lot for my scenario and it doesn't affect any existed features.

[kraih]

Could you elaborate a bit? I've brought up the topic on IRC and we were having trouble understanding the use case.

[kraih]

$fh is initialized but not used.

[iakuf]

I was in China, can not connect to the IRC server.

[iakuf]

I am here to develop a HTML5 file upload to a binary file into a way to upload the data block .
Client upload 1M, it will return the current size per upload to upload, and then disconnects.After re-create the connection from the position of the file transfer completed last start transferring files.
So I have come to the file related operations by Mojo::Asset::File module. in this way, users can upload breakpoint resume. the process can be interrupted at any, some time after resuming.

[iakuf]

Much like the following, each connection, open a file, and then write additional blocks.
hook after_build_tx => sub {
my $tx = shift;
my $app = shift;
weaken $tx;

$tx->req->content->on(body => sub {
    my $single  = shift;
    return unless $tx->req->method eq 'POST' or $tx->req->method eq 'OPTIONS';
    return unless $tx->req->url->path->contains('/upload');

    my $file = Mojo::Asset::File->new(path => $tx->req->param("path"));
    $single->unsubscribe('read')->on(read => sub {
        my ($single, $bytes) = @_; 
        $file->add_chunk($bytes);
    }); 
}); 

};
After completion of the transfer will return the size of 1M.
any [qw(POST OPTIONS GET)] => '/upload' => sub {
my $self = shift;
my $path = $self->param('path');

my $filesize = -f $path ? -s $path : 0 ;
return $self->render(json => {
    start   => $filesize,
    success => 1,
    message => '',
});

};
If you have not finished 1M transfer, only passing a portion of 1M, this "get / upload" interfaces can also get the results of the current size then the client can then resume at that point.

[iakuf]

We're sorry, the second paragraph of the code is wrong.

[kraih]

Thanks, applied with a few small changes.

[iakuf]

Thank you very much for supporting me, my html5 large files "breakpoint resume" developed using the latest Mojolicous finished.

My project addresses: https://github.com/iakuf/mojolicious-stream-upload

[kraih]

I'm afraid this is going to be reverted in Mojolicious 5.52, because there have been too many problems with read-only file systems. Lots of test failures from CPAN Testers and user complaints. 04aa323

[kiwiroy]

Is there any support for using utime(undef, undef, $path) in place of -w _ as a means of ascertaining write-able in that test?