Strip subresource hints

[GoogleCodeExporter]

mod_pagespeed is not aware of subresource links, so doesn't correct them.

Original issue reported on code.google.com by keylocat...@gmail.com on 18 Aug 2014 at 6:58

[jeffkaufman]

This is tricky. Imagine you have a page:

<link rel=subresource src=foo>
...
<img src=foo>

Then PageSpeed changes foo to foo.pagespeed.ic.HASH.jpg in the img but not the subresource. That's really bad, because it makes the subresource tag misleading to the browser, actively harmful.

Unfortunately we may not be able to correctly rewrite the subresource tag, because what url we use depends on context. For example, we might have resized an image, or combined some resources.

To fix this we need to:

• Remove any existing subresource tags/headers
• Add new ones using the property cache when we think it's worth it (even if the page doesn't have them already)

We should do the first part now, because we're making things worse. The second one is substantial work, and is more of a new feature, but would still be good.

I'm opening #1158 for the feature, and keeping this for the bug.

[oschaaf]

I'll take a look at stripping out any existing subresource tags/headers

[jmarantz]

I think we should use the pcache for this. If we strip all subresources
blindly we may remove ones that pagespeed will not be rewriting. In fact I
expect that would be a common case because subresources referenced from js
would benefit most from hints since the preloaded scanner would miss those.
On Oct 26, 2015 8:53 AM, "Otto van der Schaaf" notifications@github.com
wrote:

I'll take a look at stripping out any existing subresource tags/headers

—
Reply to this email directly or view it on GitHub
#973 (comment)
.

[jeffkaufman]

@jmarantz Yes, I agree we should use the property cache, and I've filed #1158 for that.

If people are mostly using subresource tags to reference things loaded by js then we should leave existing ones in for now, while if they're mostly referencing things loaded declaratively then we should remove them. I'll try to find out.

[jmarantz]

IMO there is not a strong reason to use subresource hints for resources
ref'd in html, as they can be pre-loaded. The only exception is when the
HTML starts streaming thru before (say) DB lookups are done.

On Mon, Oct 26, 2015 at 9:34 AM, Jeff Kaufman notifications@github.com
wrote:

@jmarantz https://github.com/jmarantz Yes, I agree we should use the
property cache, and I've filed #1158
#1158 for that.

If people are mostly using subresource tags to reference things loaded by
js then we should leave existing ones in for now, while if they're mostly
referencing things loaded declaratively then we should remove them. I'll
try to find out.

—
Reply to this email directly or view it on GitHub
#973 (comment)
.

[jeffkaufman]

I just ran a search for examples of pages with rel=subresource. I found subresource links for resources that were loaded:

• 12 pages: js referenced from inline js
• 11 pages: js referenced directly from page
• 5 pages: css referenced directly from page
• 3 pages: image referenced from page
• 2 pages: json referenced from js referenced from page
• 2 pages: css referenced directly from page before the subresource hint
• 2 pages: js not referenced from anywhere (one was old version of show_ads_impl)
• 1 page: font referenced from css referenced from page
• 1 page: font referenced from js referenced from page
• 1 page: font not referenced from anywhere
• 1 page: image (loading gif) referenced from js referenced from page
• 1 page: swf not loaded by page
• 1 page: css not referenced from anywhere (print stylesheet)
• 1 page: image referenced from inline css before the subresource hint
• 1 page: image referenced from inline js
• 1 page: css referenced from inline js

(When multiple pages were generated from one template I only counted that once.)

First, a bunch of these (7) are just making the page worse and it would be great to strip them out:

• 2 pages: css referenced directly from page before the subresource hint
• 2 pages: js not referenced from anywhere (one was old version of show_ads_impl)
• 1 page: swf not loaded by page
• 1 page: image referenced from inline css before the subresource hint
• 1 page: css not referenced from anywhere (print stylesheet)

Of the remaining 39, PageSpeed currently would rename the referenced resource (making the rel=subresource harmful) 21 times and leave it alone the other 18.

So there are cases where there are useful hints that we would strip, but stripping is better than not stripping right now.

[jeffkaufman]

If PreserveUrls is on we shouldn't strip, though.

[jeffkaufman]

Do we have consensus that we should:

• strip these for now unless PreserveUrls is on
• medium term, write a filter to add these (Insert subresource hints #1158)

?

[oschaaf]

@jeffkaufman That makes sense to me