Digital preservation

For now, this will be somewhat of a backup of my Quora blog on the topic, which is now not being maintained. (I’m trying to integrate content from here to more fitting pages.)

Thoughts on storing information in a useful/easily-accessible way. Archiving, backups, single-source publishing, source code management, redundancy (local, cloud), link rot, etc. Essentially, how can we best store thoughts so that later (a day? month? years?) we can easily find them again.

High-level overview of strategies for archiving browser activity

• Using a download program external to the browser. This is what gwern does; he uses wget to download a list of URLs. The list of URLs are obtained from places.sqlite, which Firefox generates.

• Using a browser plugin to save the page content to disk. The idea of this one is: You already have the webpage content in memory by virtue of browsing the page, so why not just store that to disk? However, this turns about to be harder than I thought, because first you need to understand JavaScript, and then you need to understand the plugin architecture of each major browser, and then actually write a plugin. It doesn’t seem like there are any GOOD plugins out there. Shelve is one option but I don’t trust it.

• Using a cache server and then tunneling all your browser activity through a proxy provided by that server. I think Squid is the dominant player here, but I haven’t much experience with this strategy in general. One advantage of this is, you can set it up once, and it will work for all browsers (that can tunnel through a proxy).

• Browser cache? Firefox and Chrome both maintain a cache of pages. The question is, does this count as a cache server, or serialized browser DOM, or a completely different solution?

Here is a table summarizing the strategies in more detail. The “Type” column is defined as follows:

External

A process external to the browser is run, and fetches pages separately. The advantages of this approach are that it is guaranteed to run asynchronously and it is easy to automate. One disadvantage is that all content must be downloaded twice (once when viewing the page in the browser, and once when archiving the page through the external process). Another is that the effects of scripting are generally not present.

Browser DOM

The browser fetches the pages, and internally represents it as a DOM. This is a serialized form of the DOM. The advantages of this approach are that the content need only be downloaded once, and the effects of scripting are generally(?) present.

Raw

The browser fetches the pages, but a process intercepts the streams reaching it to store it for the long-term. One advantage: theoretically, every bit that is transmitted is captured. Disadvantages: the way things are cached may be different from how the browser represents the page as a DOM(?); effects of client-side scripting are not present(?).

Source HTML vs generated HTML

Thanks to JavaScript and other client-side scripting, naively saving the source HTML of a webpage is sometimes not enough to save all the content.

There is a concept of “generated HTML”, which is the HTML that results from applying scripts to the source HTML.

Is there a reliable way to access this generated HTML?

Separately (?) there is the concept of the browser DOM, which is the object-oriented representation of the page, as interpreted by the browser. The DOM acts as an API that languages like JavaScript can use. The browser DOM differs from the source HTML is at least two respects:

• It corrects for “in the wild” HTML by e.g. completing tags that were not closed in the original HTML.
• It applies scripts to the original HTML.

Does the DOM differ from the “generated HTML”?

This is a good read: Best Way to View Generated Source of Webpage?.

What are various things that can go wrong?

• JavaScript alters source HTML
• JavaScript loads more content as you interact with the page (e.g. “See more” and replies in Facebook, Quora, etc.)
• Login required

See https://exp.issarice.com/archive_test/ for several tests to run on archival programs.

Facebook.com DOM

I downloaded https://facebook.com while logged in, after scrolling down on my newsfeed a couple of times and expanding some things. I downloaded the page using three methods:

• document.documentElement.outerHTML
• DOM Inspector plugin
• The Web Developers plugin’s “generated HTML” feature

To diff the files, I first pre-processed the files using fmt -w 40 and tidy -utf8 and then did the actual comparison using gvimdiff. Passing -w 40 to fmt is so that three files could comfortably sit side-by-side on my screen. gvimdiff is convenient in that it allows for three files to be compared at once.

DOM Inspector uses things like

<a class="see_more_link" onclick='var func = function(e) {
e.preventDefault(); }; var parent = Parent.byClass(this,
"text_exposed_root") ...' ...>...</a>

On the other hand, the bookmarklet and Web Developer Tools “generated HTML” do things like

<a class="see_more_link" onclick="var func = function(e) {
e.preventDefault(); }; var parent = Parent.byClass(this,
&quot;text_exposed_root&quot;) ..." ...>...</a>

The bookmarklet also compresses contiguous spaces into a single space, and breaks lines on its own.

In other words, some differences are merely due to different ways of serializing the same underlying structure.

Here is another. On the bookmarklet and Web Dev tools:

<abbr class="livetimestamp"
title="Monday, October 31, 2016 at
2:55am" data-utime="1477907708"
data-shorten="true">8 hrs</abbr>

On the DOM Inspector:

<abbr class="livetimestamp"
title="Monday, October 31, 2016 at
2:55am" data-utime="1477907708"
data-shorten="true"><span class="timestampContent">8 hrs</span></abbr>

Note that here, the structure itself is different. There is no span tag with class timestampContent in the former.

There are also some positional stuff that is different, e.g. chat contact list? Some of it could be because the pages were not captured at exactly the same time.

Some div tags corresponding to people (in the chat bar?) were also missing in one or more of the dumps.

The Web Dev tools version also didn’t have some script tags that referenced external scripts.

However, as far as I can tell, the actual content on the page is stored in all versions. Indeed, du -h * shows the same size (1.2M) for all versions so the differences are a rounding error. Given this, it seems like the easiest way to programmatically access the current browser DOM is through document.documentElement.outerHTML.

Various bookmarklets

You will have to run these through a JavaScript compressor like yui-compressor.

// Open a new window and dump the current DOM. This can then be saved with
// "Save Page As".
javascript:(function() {
  var para = document.createElement('p');
  var t = document.createTextNode(document.documentElement.outerHTML.toString());
  para.appendChild(t);
  var win = window.open();
  win.document.body.appendChild(para);
})();

There is a variant to the above that uses win.document.write() and win.document.close() instead of manipulating the DOM with methods like appendChild(). However, I found that this fails on some sites for some unknown reason. One page it fails on is https://developer.mozilla.org/en-US/docs/Using_files_from_web_applications. Another is http://www.tecmint.com/best-linux-file-diff-tools-comparison/.

From a Stack Overflow answer:

document.write is one of the oldest vestiges of ancient JavaScript, and should generally be avoided. Instead, you probably want to use DOM [manipulation] methods to update the document.

I don’t know enough about JavaScript or its history to know if this is true, but it’s parsimonious.

For small pages, doing

// Pick one
copy(document.documentElement.outerHTML);
copy(document.body.innerHTML);

is another option. This copies the DOM to your clipboard. I found that with very large sites, the clipboard seems to get full and not update. See http://stackoverflow.com/a/20023875/3422337.

Dissecting ScrapBook X

Git tree that I am using.

ScrapBook X seems to improve ScrapBook, and is apparently based on it. I couldn’t find the source for the original ScrapBook, so I can’t really see if this is true.

Anyway, my general intent is to get an idea of the high-level strategy it uses, and then see if there is a way to save things without blocking the browser.

Some notes follow.

chrome/content/scrapbook/ seems to be where the meat of the plugin is stored. Here, I first noticed capture.js, but there is actually also a saver.js, which turned out to be where the actual work is being done.

Reading capture.js, I see var sbCaptureTask. Reading in, I see _captureWindow: function(aWindow, aAllowPartial) on line 470. This calls

var ret = gContentSaver.captureWindow(aWindow, aAllowPartial, gShowDetail, gResName, gResIdx, preset, gContext, gTitles[this.index]);

on line 491. But line 2 defines var gContentSaver = new sbContentSaverClass();. This is the only occurrence of sbContentSaverClass in this file. No worry, I run

:vimgrep /sbContentSaverClass/ **/*.js

and find

chrome/content/scrapbook/capture.js|2 col 25| var gContentSaver = new sbContentSaverClass();
chrome/content/scrapbook/saver.js|4 col 25| var saver = new sbContentSaverClass();
chrome/content/scrapbook/saver.js|9 col 25| var saver = new sbContentSaverClass();
chrome/content/scrapbook/saver.js|42 col 10| function sbContentSaverClass() {
chrome/content/scrapbook/saver.js|88 col 1| sbContentSaverClass.prototype = {

So the class is in saver.js. Line 42 looks promising (the fourth result above). We want captureWindow, so search downward. Lines 214–216:

// save the document content to ScrapBook
this.contentDir = sbCommonUtils.getContentDir(this.item.id);
var newName = this.saveDocumentInternal(aRootWindow.document, this.documentName);

So now we want saveDocumentInternal, and it looks like aRootWindow.document is what is being saved, which should be analogous to document if the entire page is sought. In Vim [I looks for all occurrences of the keyword under the cursor, so with the cursor in saveDocumentInternal, I hit [I. The definition is on line 272.

Several interesting things here. First off,

if ( ["text/html", "application/xhtml+xml"].indexOf(contentType) < 0 ) {
    if ( !(aDocument.documentElement.nodeName.toUpperCase() == "HTML" && this.option["fileAsHtml"]) ) {
        captureType = "file";
    }
}
if ( captureType ) {
    var newLeafName = this.saveFileInternal(aDocument.location.href, aFileKey, captureType, charset);
    return newLeafName;
}

So it looks like if the root node is HTML, we call saveFileInternal. But this looks confusing because we’re only passing aDocument.location.href rather than aDocument or aDocument.documentElement, which contains the DOM! But the test

if ( ["text/html", "application/xhtml+xml"].indexOf(contentType) < 0 ) {

seems to be checking whether the content type is in the list. Regular HTML files should have text/html as the content type, so it seems that this capture only happens when the content type is not HTML yet the root element is HTML. I’m not sure why we would make this a special case.

Going past the block and reading down, line 316 has:

var htmlNode = aDocument.documentElement;

And then line 394:

if ( !this.selection ) {
    rootNode = htmlNode.cloneNode(true);

So rootNode is now a copy of htmlNode; tracking both and scrolling down, line 525 has:

myHTML += sbCommonUtils.surroundByTags(rootNode, rootNode.innerHTML + "\n");

And finally line 535:

sbCommonUtils.writeFile(myHTMLFile, myHTML, charset);

So it is myHTML that gets written in the end. But myHTML takes from rootNode.innerHTML, which is htmlNode.innerHTML, which is aDocument.documentElement.innerHTML.

To summarize, it seems that ScrapBook X (and hence probably ScrapBook also) saves the browser DOM by accessing document.documentElement.innerHTML. Of course, the plugin does a lot more because it also tries to fetch dependencies like stylesheets and various media.

I’m still not sure why ScrapBook X blocks Firefox while it saves a page. I think plugins are allowed to fork separate processes, so it should be possible to run the archival in the background while the user is free to do whatever else with Firefox.

Naming archives

How do you name archives? In general, using the URL is not possible, because there is around a 2000-character limit to URLs but file systems like ext4 have a 255-byte limit.

Solution in ELinks

ELinks comes with Lua scripting support. Since it provides a hook that runs when the page loads and gives you the URL and HTML, the solution is pretty simple:

-- From http://stackoverflow.com/a/4991602
function file_exists(name)
    local f = io.open(name, "r")
    if f ~= nil then
        io.close(f)
        return true
    else
        return false
    end
end

function pre_format_html_hook(url, html)
    local timestamp = os.date("%Y-%m-%dT%H-%M-%S%z")
    -- Increment i until we get a fresh location we can write to. This is
    -- necessary not because someone can open multiple pages in the same
    -- second, but rather because ELinks occasionally downloads auxiliary
    -- files, such as CSS files, as it fetches the page.
    local i = 0
    while file_exists(timestamp .. "." .. i .. ".html") do
        i = i + 1
    end
    local name = timestamp .. "." .. i .. ".html"
    -- After obtaining the HTML page, save a copy of it timestamped
    local index = io.open("index.txt", "a")
    index:write(name .. "\t" .. url .. "\n")
    index:close()
    local file = io.open(name, "w")
    file:write(html)
    file:close()
    -- Return the HTML page unmodified
    return nil
end

Just save it in ~/.elinks/hooks.lua.

Question: Does this capture everything that can be seen in ELinks? In other words, is there a one-to-one correspondence between ‘things that can be see when browsing the page’ and ‘things that are saved on disk’? Since Elinks doesn’t support JavaScript, this seems possible, but I do wonder if e.g. the content of iframes are saved (not that many sites I browse use these). Periodically grepping around the archived content is probably a good idea to see if anything is missing.

Also the ELinks handling of HTTPS connections (as of version 0.12pre6 on Ubuntu) is a bit concerning.

Inactive account policy of various services

I’d like to do something similar for other services to what I did for Dropbox in a Quora answer.

The internet is so young that inactive accounts due to death has not been that big of a deal yet. Legal deposit laws, libraries, etc., have kept old books from disappearing, but what can we say about internet content?

Requirements for good data archiving solutions

Some ideas for now:

• Pandoc-flavored markdown
• LaTeX, then use LaTeX2WP or LaTeX–HTML converters (the benefit of this is that printing is covered, as long as your LaTeX skills are decent)
• raw HTML (no)
• using different formats for different content may be possible, e.g. using markdown for most content, but switching to LaTeX for math-heavy content.

Static page generation

• Static Site Generators —complete(?) list
• Static site generators
• Jekyll • Simple, blog-aware, static sites
• Hakyll - Home
• Switching from Jekyll to Hakyll
• The Switch to Hakyll

Best Jekyll doc seems to be the video available from here: algonquindesign/jekyll. Okay, so I’ve finally figured out how to use jekyll in conjunction with Github. My test website is up on Hello world. See riceissa/abc for the source. The trickiest part was modifying all of jekyll’s default internal links (e.g. automatically generated links to blog posts) to use {{site.baseurl}}, so that it worked both locally and through Github pages.

I’m still trying to figure out how to do math properly through jekyll, because at the moment I must use two backslashes, since one gets eaten up by Markdown (pandoc can prevent this, so either switch to hakyll or use a plugin for jekyll?—does Github even allow plugins for jekyll?)

More on link rot and the Internet Archive (EDIT: found the speech!!)

Ones with a star (*) are more useful.

• Why URL shorteners suck
• URL shorteners suck less, thanks to the Internet Archive and 301Works
• URLTeam - Archiveteam
• URLTE.AM
• Learn From History: Why The Internet Archive Is So Important
• Brewster’s trillions: Internet Archive strives to keep web history alive
• on url shorteners
• 301works-faq : Free Download & Streaming : Internet Archive

I’m still trying to find that speech about link rot and URL shorteners.

EDIT: I found the speech: The Splendiferous Story of Archive Team. The guy’s name is Jason Scott Sadofsky and he seems to have a lot of stuff e.g. T E X T F I L E S. He actually seems to be part of the Archiveteam, which makes him even cooler. (How did I find the speech again after two days of trying? [So that I can get better at finding old things.] Through this article: This Group Is Saving The Web From Itself (And Rescuing Your Stuff), where I got his name. My search terms on Google to find that article were “saving the internet archive”; after I recognized the name, I just Googled “jason scott”, and found his Wikipedia page and website, whose design I remembered from when I read the speech before.)

The relevant bit on URL shorteners is:

URL shorteners may be one of the worst ideas, one of the most backward ideas, to come out of the last five years. In very recent times, per-site shorteners, where a website registers a smaller version of its hostname and provides a single small link for a more complicated piece of content within it.. those are fine. But these general-purpose URL shorteners, with their shady or fragile setups and utter dependence upon them, well. If we lose TinyURL or bit.ly, millions of weblogs, essays, and non-archived tweets lose their meaning. Instantly.

Terminally Incoherent

from Luke.

• Personal Backups
• Putting your Vim files under version control
• You Need Backups
• Backing up your work: Common Sense 101
• Maybe the Backup Problem Will Resolve Itself In Time
• Backup is not just for geeks

The message is clear.

Also example of backing up dotfiles: maciakl/.dotfiles. Somewhere I remember reading a post about how to set this up nicely (by Luke), but can’t seem to locate it at the moment.

Gwern on archiving

Archiving URLs see also http://www.gwern.net/About#long-content. I can’t believe I had forgotten about this! Okay and linked on the first page is this: https://web.archive.org/web/20141208192153/http://www.gwern.net/docs/2011-muflax-backup.pdf (from muflax).

URL shortening

where is the speech that someone at Digital Library of Free Books, Movies, Music & Wayback Machine made about how URL shortening sites cause a lot of link rot?

EDIT: Okay I found this at least: http://www.archiveteam.org/index.php?title=URLTeam and this links to Why URL shorteners are bad. Not bad.

EDIT2: Okay finally found it. See More on link rot and the Internet Archive (EDIT: found the speech!!).

What do people do about archiving information that they don’t own?

e.g. a journal article that you didn’t write. This sort of thing could be very useful for people to have access to. Storing them locally wouldn’t be a problem, and putting them up online shouldn’t cause too much trouble either. Websites like http://archive.org always just archive whatever they can find anyway.

Actually, how does Digital Library of Free Books, Movies, Music & Wayback Machine deal with copyright issues?

Git/GitHub

See What are some unconventional and unique uses of Git? and also the article linked in the description of the question (http://www.wired.com/wiredenterprise/2012/02/github-revisited/).

Git in particular is excellent in terms of having a local mirror as well as one online (e.g. by using Github/Bitbucket), along with all of the changes that have been made. This makes the data redundant/safe. It doesn’t seem to be bad either in terms of sharing, since everything will be in plaintext. Posting on Quora might be a problem though, since it doesn’t use Markdown or the like.

I suppose the other big problem with using source-control is the question of storing binary data. Use a different place to store those? (e.g. putting all plaintext on Github but uploading photos to Wordpress, and linking them?) Deal with binaries in git as well, and make many small projects so that it doesn’t slow down git?

Random questions (not Quora-quality yet)

Some of these have probably already been asked, so.

• Is single-sourcing worth one’s time for most things? For some things?
• How important is grammar/capitalization for recording most thoughts? (cf AKC’s post [on wordpress?] about grammar not being important)
• Is Evernote reliable?
• How do I easily create multiple mirrors of my data, both locally and online?
• How do I reliable backup online data?
• How do I keep a revision list of online data?—version control it somehow? how?
• Why is Quora’s search feature unreliable at finding the information I want? e.g. it can’t find comments.
• Why does Quora make everything so hard to find? —e.g. there are no chronological lists of all posts of a user, no place to find all the posts I’ve upvoted, and so on.
• How do I make sharing easier between Quora and other sites?

External mirroring

The way I see it, external page savers like the Internet Archive and WebCite shouldn’t be trusted; rather, they’re a convenient way to avoid copyright violations (since they’re hosting the files, not you); and they provide a time buffer for you to get local copies.

I don’t think I should be surprised at all by this, but I was quite impressed with Perma.cc’s contingency plan, which seems like an obvious improvement over standard web services that simply shut down and give no notice¹.

Archive buttons

On Firefox, I have the following archive buttons:

From left to right, these are Pinboard, Zotero, archive.is, and Archive.org. The Archive.org button is a bookmarklet with the code:

javascript:void(window.open('https://web.archive.org/save/'+location.href))

I also have the Scrapbook plugin for Firefox. This allows me to quickly create copies that are local (Scrapbook and Zotero), private network (Pinboard), and public network (Archive.org).

See also

• Archiving Facebook

Backlinks

• single source publishing.md