For context, this is the Lars Ingebrigtsen who wrote the manual for Gnus[0], a common Emacs package for reading email and Usenet. It’s clever, funny, and wildly informative. Lars has probably forgotten more about email parsing than 99% of us here will ever have learned.
The manual itself says[1]:
> Often when I read the manual, I think that we should take a collection up to have Lars psycho-analysed.
The real punchline is that this is a perfect example of "just enough knowledge to be dangerous." Whoever processed these emails knew enough to know emails aren't plain text, but not enough to know that quoted-printable decoding isn't something you hand-roll with find-and-replace. It's the same class of bug as manually parsing HTML with regex, it works right up until it doesn't, and then you get congressional evidence full of mystery equals signs.
> It's the same class of bug as manually parsing HTML with regex, it works right up until it doesn't
I'm sure you already know this one, but for anyone else reading this I can share my favourite StackOverflow answer of all time: https://stackoverflow.com/a/1732454
I prefer the question about CPU pipelines that gets explained using a railroad switch as example. That one does a decent job of answering the question instead of going of on a, how to best put it, mentally deranged one page rant about regexes with the lazy throw away line at the end being the only thing that makes it qualify as an answer at all.
The regex answer is from the very old days of Stackoverflow, before fun was banned. I agree it barely qualifies as answer, but considering that the question has over 4 million page views (which almost puts it in the top 100 most viewed questions all-time), it has reached a lot people. The answer probably had much more influence than any serious answer on that topic. So I'd say the author did a good job.
I know it's a hassle for a platform to moderate good rants from bad ones, and I decry SO from pushing too hard against these. I truly believe that our industry would benefit from more drunken technical rants.
People have shared it here and on reddit a bunch of times because it's funny. I always found the pragmatic counter-answer about using regex and the comments about how brittle it is to parse XML properly assuming a specific structure to be much more useful.
It took me years to notice, but did you catch that the answer actually subtly misinterprets what the question is asking for?
Guy (in my reading) appears to talk about matching an entire HTML document with regex. Indeed, that is not possible due to the grammars involved. But that is not what was being asked.
What was being asked is whether the individual HTML tags can be parsed via regex. And to my understanding those are very much workable, and there's no grammar capability mismatch either.
I know this is grumpy but this I’ve never liked this answer. It is a perfect encapsulation of the elitism in the SO community—if you’re new, your questions are closed and your answers are edited and downvoted. Meanwhile this is tolerated only because it’s posted by a member with high rep and username recognition.
As someone who used to write custom crawlers 20 years ago, I can confirm that regular expressions worked great. All my crawlers were custom designed for a page and the sites were mostly generated by some CMS and had consistent HTML. I don't remember having to do much bug fixes that were related to regular expression issues.
I don't suggest writing generic HTML parsers that works with any site, but for custom crawlers they work great.
Not to say that the tools available are the same now as 20 years ago. Today I would probably use puppeteer or some similar tool and query the DOM instead.
An interesting thing is that most webpages are generated using text templates. There's some text processing like escaping special characters, but it's mostly text that happened to be (somewhat) valid HTML.
So extracting information from this text with regexps often makes perfect sense.
I would distinguish between parsing and scraping. Parsing really needs a, well, parser. Otherwise you’ll get things wrong on perfectly well formed input and your program will be brittle and weird.
A scraper is already resigned to being brittle and weird. You’re relying not only on the syntax of the data, but an implicit structure beyond that. This structure is unspecified and may change without notice, so whatever robustness you can achieve will come from being loose with what you accept and trying to guess what changes might be made on the other end. Regex is a decent tool for that.
Funny how differently people can perceive things. That's my least favorite SO answer of all time, and I cringe every time I see it.
It's a very bad answer. First of all, processing HTML with regex can be perfectly acceptable depending on what you're trying to do. Yes, this doesn't include full-blown "parsing" of arbitrary HTML, but there are plenty of ways in which you might want to process or transform HTML that either don't require producing a parse tree, don't require perfect accuracy, or are operating on HTML whose structure is constrained and known in advance. Second, it doesn't even attempt to explain to OP why parsing arbitrary HTML with regex is impossible or poorly-advised.
The OP didn't want his post to be taken over by someone hamming it up with an attempt at creative writing. He wanted a useful answer. Yes, this answer is "quirky" and "whimsical" and "fun" but I read those as euphemisms for "trying to conscript unwilling victims into your personal sense of nerd-humor".
The whole argument hinges on one word in your post: arbitrary.
I parse my own HTML I produce directly in a context where I fully control the output. It works fine, but parsing other people’s HTML is a lesson in humility. I’ve also done that, but I did it as a one time thing. I parsed a specific point in time, refusing to change that at any point.
It also hinges on another word: parsing. There are things other than parsing that you might want to do. For example, if you want to count the number of `<hr>` tags in an HTML document, that doesn't require parsing it, and can indeed be done with regex.
No you can’t. You can have an unescaped <hr> inside a script tag, for example. The best you can do is a simple string search for “<hr>” and hope it’s returning what you think it might be returning. Regexps are not powerful enough to determine whether any particular instance of “<hr>” is actually an HTML tag.
Like, it’s not a matter of cleverness, either. You can’t code around it. It’s simply not possible.
The most interesting thing to me wasn't the equals signs, which I knew are from quoted-printable, but the fact that when an equals sign appears, a letter that should have been preceding or following it is missing. It's as if an off-by-one error has occurred, where instead of getting rid of the equals sign, it's gotten rid of part of the actual text. Perhaps the CRLF/LF thing is part of it.
SMTP is a line–based protocol, including the part that transfers the message body
The server needs to parse the message headers, so it can't be an opaque blob. If the client uses IMAP, the server needs to fully parse the message. The only alternative is POP3, where the client downloads all messages as blobs and you can only read your email from one location, which made sense in the year 2000 but not now when everyone has several devices.
POP3 is more for reading and acting on your email in one place (taking notes, plan actions, discard and delete,…). No need to consume them on other devices as you’ve already extracted the important bits.
I use imap on my mobile device, but that’s mostly for recent emails until I get to my computer. Then it’s downloaded and deleted from the server.
Mails are (or used to be) processed line-by-line, typically using fixed-length buffers. This avoids dynamic memory allocation and having to write a streaming parser. RFC 821 finally limited the line length to at most 1000 bytes.
Given a mechanism for soft line breaks, breaking already at below 80 characters would increase compatibility with older mail software and be more convenient when listing the raw email in a terminal.
This is also why MIME Base64 typically inserts line breaks after 76 characters.
In early days, many/most people also read their email on terminals (or printers) with 80-column lines, so breaking lines at 72-ish was considered good email etiquette (to allow for later quoting prefix ">" without exceeding 80 characters).
I don't think kids today realize how little memory we had when SMTP was designed.
For example, the PDP-11 (early 1970s), which was shared among dozens of concurrent users, had 512 kilobytes of RAM. The VAX-11 (late 1970s) might have as much as 2 megabytes.
Programmers were literally counting bytes to write programs.
"BITNET was a co-operative university computer network in the United States founded in 1981 by Ira Fuchs at the City University of New York (CUNY) and Greydon Freeman at Yale University."
BITNET connected mainframes, had gateways to the Unix world and was still active in the 90s. And limited line lengths … some may remember SYSIN DD DATA … oh my goodness …
This is how email work(ed) over smtp. When each command was sent it would get a '200'-class message (success) or 400/500-class message (failure). Sound familiar?
Yep! And also, if you included a blank line and then the headers for a new email in the bottom of your message, you could tell the server, hey, here comes another email for you to process!
If you were typing into a feedback form powered by something from Matt’s Script Archive, there was about a 95% chance you could trivially get it to send out multiple emails to other parties for every one email sent to the site’s owner.
Back in 80s-90s it was common to use static buffers to simplify implementation - you allocate a fixed size buffer and reject a message if it has a line longer than the buffer size. SMTP RFC specifies 1000 symbols limit (including \r\n) but it's common to wrap around 87 symbols so it is easy to examine source (on a small screen).
The simplest reason: Mail servers have long had features which will send the mail client a substring of the text content without transferring the entire thing. Like the GMail inbox view, before you open any one message.
I suspect this is relevant because Quoted Printable was only a useful encoding for MIME types like text and HTML (the human readable email body), not binary (eg. Attachments, images, videos). Mail servers (if they want) can effectively treat the binary types as an opaque blob, while the text types can be read for more efficient transfer of message listings to the client.
As far as I can remember, most mail servers were fairly sane about that sort of thing, even back in the 90’s when this stuff was introduced. However, there were always these more or less motivated fears about some server somewhere running on some ancient IBM hardware using EBCDIC encoding and truncating everything to 72 characters because its model of the world was based on punched cards. So standards were written to handle all those bizarre systems. And I am sure that there is someone on HN who actually used one of those servers...
RFC822 explicitly says it is for readability on systems with simple display software. Given that the protocol is from 1982 and systems back then had between 4 and 16kb RAM in total it might have made sense to give the lower end thin client systems of the day something preprocessed.
Also it is an easy way to stop a denial of service attack. If you let an infinite amount in that field. I can remotely overflow your system memory. The mail system can just error out and hang up on the person trying the attack instead of crashing out.
Keep in mind that in ye olden days, email was not a worldwide communication method. It was more typical for it to be an internal-only mail system, running on whatever legacy mainframe your org had, and working within whatever constraints that forced. So in the 90s when the internet began to expand, and email to external organizations became a bigger thing, you were just as concerned with compatibility with all those legacy terminal-based mail programs, which led to different choices when engineering the systems.
Are you certain? Not OP, but a huge chunk of early RFCs was about how to let giant IBM systems talk to everyone else, specifying everything from character sets (nearly universally “7-bit ASCII”) to end of line/message characters. Otherwise, IBM would’ve tried to make EBCDIC the default for everything.
For instance, consider FTP’s text mode, which was primarily a way to accidentally corrupt your download when you forgot to type “bin” first, but was also handy for getting human readable files from one incompatible system to another.
I wrote my own email archiving software. The hardest part was dealing with all the weird edge cases in my 20+ year collection of .eml files. For being so simple conceptually, email is surprisingly complicated.
> So what’s happened here? Well, whoever collected these emails first converted from CRLF (i.e., “Windows” line ending coding) to “NL” (i.e., “Unix” line ending coding). This is pretty normal if you want to deal with email. But you then have one byte fewer:
I think there is a second possible conclusion, which is that the transformation happened historically. Everyone assumes these emails are an exact dump from Gmail, but isn't it possible that Epstein was syncing emails from Gmail to a third party mail server?
Since the Stackoverflow post details the exact situation in 2011, I think we should be open to the idea that we're seeing data collected from a secondary mail server, not Gmail directly.
Do we have anything to discount this?
(If I'm not mistaken, I think you can also see the "=" issue simply by applying the Quoted-Printable encoding twice, not just by mishandling the line-endings, which also makes me think two mail servers. It also explains why the "=" symbol is retained.)
In one of the email PDFs I saw an XML plist with some metadata that looked like it was from Apple's Mail.app, so these might be extracted from whatever internal format that uses.
Yeah, I wouldn't bet on this being a single bad Gmail export; it smells much more like the accumulated scars of multiple mail systems doing "helpful" things to the same messages over time
(The title of the blog reminded me the late Bob Pease [1] who had the signature, "What's all this XXX stuff, anyhow?" [2] where XXX might be "noise gain", "capacitor leakage"…)
Yeah, and using two bytes for a single line termination (or separation or whatever)? Why make things more complicated and take more space at the same time?
Remember that back in the mists of time, computers used typewriter-esque machines for user interaction and text output. You had to send a CR followed by an LF to go to the next line on the physical device. Storing both characters in the file meant the OS didn't need to insert any additional characters when printing. Having two separate characters let you do tricks like overstriking (just send CR, no LF)
True, but I don’t think there was a common reason to ever send a linefeed without going back to the beginning. Were people printing lots of vertical pipe characters at column 70 or something?
It would’ve been far less messy to make printers process linefeed like \n acts today, and omit the redundant CR. Then you could still use CR for those overstrike purposes but have a 1-byte universal newline character, which we almost finally have today now that Windows mostly stopped resisting the inevitable.
I haven't seen them other than in the submission - but if the length matches up it may be that they were processed from raw email, the RFC defines a length to wrap at.
Edit: yes I think that's most likely what it is (and it's SHOULD 78ch; MUST 998ch) - I was forgetting that it also specifies the CRLF usage, it's not (necessarily) related to Windows at all here as described in TFA.
> it's not (necessarily) related to Windows at all here as described in TFA.
The article doesn't claim that it's Windows related. The article is very clear in explaining that the spec requires =CRLF (3 characters), then mentions (in passing) that CRLF is the typical line ending on Windows, then speculates that someone replaced the two characters CRLF with a one character new line, as on Unix or other OSs.
Ok yeah I may have misinterpreted that bit in the article. It would be a totally reasonable assumption if you didn't happen to know that about email though, it wasn't a judgement regardless.
I am just wondering how it is good idea for a sever to insert some characters into user's input. If a collegue were to propose this, i d laugh in his face
It's just sp hacky i cant belive it's a real life's solution
Consider converting the original text (maintaining the author’s original line wrapping and indentation) to base64. Has anything been “inserted” into the text? I would suggest not. It has been encoded.
Now consider an encoding that leaves most of the text readable, translates some things based on a line length limit, and some other things based on transport limitations (e.g. passing through 7-bit systems.) As long as one follows the correct decoding rules, the original will remain intact - nothing “inserted.” The problem is someone just knowledgeable enough to be aware that email is human readable but not aware of the proper decoding has attempted to “clean up” the email for sharing.
Okey it does sound better from this POV. Still wierd as its a Client/UI concern, not something a server is supposed to do; whats next,adding "bold" tags on the title? Lol
SMTP is a line-oriented protocol. The server processes one line at a time, and needs to understand headers.
Infinite line length = infinite buffer. Even worse, QP is 7-bit (because SMTP started out ASCII only), so characters >127 get encoded as three bytes (equal, then two hex digits), so a 500-character non-ASCII UTF8 line is 1500 bytes.
It all made sense at the time. Not so much these days when 7-bit pipes only exist because they always have.
No, because there is a clear separation between the content and the envelop. You wouldnt expect the post office to open your physical letters and write routing instructions to the postmen for delivery
But I agree with sibling comment: it makes more sense when its called "encoding" instead of "inserting chars into original stream"
> You wouldnt expect the post office to open your physical letters and write routing instructions to the postmen for delivery
Digital communication is based on the postmen reading, transcribing and copying your letters. There is a reason why digital communication is treated differently then letters by the law and why the legally mandated secrecy for letters doesn't apply to emails.
It's called escaping, and almost every protocol has it. HN must convert the & symbol to & for displaying in HTML. Many wire protocols like SATA or Ethernet must insert a 1 after a certain number of consecutive 0s to maintain electrical balance. Don't remember which ones — don't quote me that it's SATA and Ethernet.
Same here. I did notice what I think was an actual error on someone's part, there was a chart in the files comparing black to white IQ distributions, and well, just look at it:
Me too. I first assumced it was an OCR error, then remembered they were emails and wouldn't need to go through OCR. Then I thought that the US Government is exactly the kind of place to print out millions of emails only to scan them back in again.
Great. Can't wait for equal signs to be the next (((whatever this is))). Maybe it's a secret code. j/k
On a side note: There are actually products marketed as kosher bacon (it's usually beef or turkey). And secular Jews frequently make jokes like this about our kosher bros who aren't allowed to eat the real stuff for some dumb reason like it has too many toes.
We’ve become so accustomed to modern libraries handling encoding transparently that when raw data surfaces (like in these dumps), we often lack the 'Digital Archeology' skills to recognize basic Quoted-Printable.
These artifacts (=20, =3D) are effectively fossils of the transport layer. It’s a stark reminder that underneath our modern AI/React/JSON world, the internet is still largely held together by 7-bit ASCII constraints and protocols from the 1980s.
Unicode labels U+000A as all of "LINE FEED (LF)", "new line (NL)" and "end of line (EOL)". I'm guessing different names were imported from slightly different character sets, although I understand the all-uppercase name to be the main/official one.
NL, or New Line, is a character in some character sets, like old mainframe computers. No need to be snarky just because he mistyped or uses a different name for something.
The writer presumably knows that umlauts and other non-ascii characters are functional in many languages. "rock döts" is poking fun at the trend in a certain tranche of anglophone rock/metal to use them in a purely aesthetic way in band names etc.
For context, this is the Lars Ingebrigtsen who wrote the manual for Gnus[0], a common Emacs package for reading email and Usenet. It’s clever, funny, and wildly informative. Lars has probably forgotten more about email parsing than 99% of us here will ever have learned.
The manual itself says[1]:
> Often when I read the manual, I think that we should take a collection up to have Lars psycho-analysed.
0: https://www.gnu.org/software/emacs/manual/html_mono/gnus.htm...
1: https://www.gnus.org/manual.html
The real punchline is that this is a perfect example of "just enough knowledge to be dangerous." Whoever processed these emails knew enough to know emails aren't plain text, but not enough to know that quoted-printable decoding isn't something you hand-roll with find-and-replace. It's the same class of bug as manually parsing HTML with regex, it works right up until it doesn't, and then you get congressional evidence full of mystery equals signs.
> It's the same class of bug as manually parsing HTML with regex, it works right up until it doesn't
I'm sure you already know this one, but for anyone else reading this I can share my favourite StackOverflow answer of all time: https://stackoverflow.com/a/1732454
I prefer the question about CPU pipelines that gets explained using a railroad switch as example. That one does a decent job of answering the question instead of going of on a, how to best put it, mentally deranged one page rant about regexes with the lazy throw away line at the end being the only thing that makes it qualify as an answer at all.
The regex answer is from the very old days of Stackoverflow, before fun was banned. I agree it barely qualifies as answer, but considering that the question has over 4 million page views (which almost puts it in the top 100 most viewed questions all-time), it has reached a lot people. The answer probably had much more influence than any serious answer on that topic. So I'd say the author did a good job.
Of all the things I wrote on SO, including many actually-useful detailed explanations, it was this drunken rant that stuck, for some reason.
I think of, and look up, this drunken rant at least once a year.
And for that I applaud you.
I know it's a hassle for a platform to moderate good rants from bad ones, and I decry SO from pushing too hard against these. I truly believe that our industry would benefit from more drunken technical rants.
People have shared it here and on reddit a bunch of times because it's funny. I always found the pragmatic counter-answer about using regex and the comments about how brittle it is to parse XML properly assuming a specific structure to be much more useful.
For anyone wondering about the railroad switch post: https://stackoverflow.com/questions/11227809/why-is-processi...
This is new to me, and a wonderful dive that I wish I was aware of during my OS course. Thanks!
But--and this is crucial--the one about regexes is hilarious.
It also comes from a time in Internet culture when humor was appreciated instead of aggressively downvoted.
It took me years to notice, but did you catch that the answer actually subtly misinterprets what the question is asking for?
Guy (in my reading) appears to talk about matching an entire HTML document with regex. Indeed, that is not possible due to the grammars involved. But that is not what was being asked.
What was being asked is whether the individual HTML tags can be parsed via regex. And to my understanding those are very much workable, and there's no grammar capability mismatch either.
I think even for single opening tags like asked there are impossible edge cases.
For example, this is perfectly valid XHTML:
I know this is grumpy but this I’ve never liked this answer. It is a perfect encapsulation of the elitism in the SO community—if you’re new, your questions are closed and your answers are edited and downvoted. Meanwhile this is tolerated only because it’s posted by a member with high rep and username recognition.
I think this answer was tolerated when SO wasn't as bad as it is now, and wouldn't be tolerated now from anyone.
It's because SO at the time was a small high-trust society where "everyone knew each other" and so things flew back then that wouldn't fly now.
As someone who used to write custom crawlers 20 years ago, I can confirm that regular expressions worked great. All my crawlers were custom designed for a page and the sites were mostly generated by some CMS and had consistent HTML. I don't remember having to do much bug fixes that were related to regular expression issues.
I don't suggest writing generic HTML parsers that works with any site, but for custom crawlers they work great.
Not to say that the tools available are the same now as 20 years ago. Today I would probably use puppeteer or some similar tool and query the DOM instead.
An interesting thing is that most webpages are generated using text templates. There's some text processing like escaping special characters, but it's mostly text that happened to be (somewhat) valid HTML.
So extracting information from this text with regexps often makes perfect sense.
I would distinguish between parsing and scraping. Parsing really needs a, well, parser. Otherwise you’ll get things wrong on perfectly well formed input and your program will be brittle and weird.
A scraper is already resigned to being brittle and weird. You’re relying not only on the syntax of the data, but an implicit structure beyond that. This structure is unspecified and may change without notice, so whatever robustness you can achieve will come from being loose with what you accept and trying to guess what changes might be made on the other end. Regex is a decent tool for that.
HE COMES
Funny how differently people can perceive things. That's my least favorite SO answer of all time, and I cringe every time I see it.
It's a very bad answer. First of all, processing HTML with regex can be perfectly acceptable depending on what you're trying to do. Yes, this doesn't include full-blown "parsing" of arbitrary HTML, but there are plenty of ways in which you might want to process or transform HTML that either don't require producing a parse tree, don't require perfect accuracy, or are operating on HTML whose structure is constrained and known in advance. Second, it doesn't even attempt to explain to OP why parsing arbitrary HTML with regex is impossible or poorly-advised.
The OP didn't want his post to be taken over by someone hamming it up with an attempt at creative writing. He wanted a useful answer. Yes, this answer is "quirky" and "whimsical" and "fun" but I read those as euphemisms for "trying to conscript unwilling victims into your personal sense of nerd-humor".
The whole argument hinges on one word in your post: arbitrary.
I parse my own HTML I produce directly in a context where I fully control the output. It works fine, but parsing other people’s HTML is a lesson in humility. I’ve also done that, but I did it as a one time thing. I parsed a specific point in time, refusing to change that at any point.
It also hinges on another word: parsing. There are things other than parsing that you might want to do. For example, if you want to count the number of `<hr>` tags in an HTML document, that doesn't require parsing it, and can indeed be done with regex.
No you can’t. You can have an unescaped <hr> inside a script tag, for example. The best you can do is a simple string search for “<hr>” and hope it’s returning what you think it might be returning. Regexps are not powerful enough to determine whether any particular instance of “<hr>” is actually an HTML tag.
Like, it’s not a matter of cleverness, either. You can’t code around it. It’s simply not possible.
There's nothing that brings joy into this world quite like the guy waiting around to tell people he doesn't like the thing they like.
And because the output still looks mostly readable, nobody questions it until years later when it's suddenly evidence in front of Congress
They have top men working on it right now.
The most interesting thing to me wasn't the equals signs, which I knew are from quoted-printable, but the fact that when an equals sign appears, a letter that should have been preceding or following it is missing. It's as if an off-by-one error has occurred, where instead of getting rid of the equals sign, it's gotten rid of part of the actual text. Perhaps the CRLF/LF thing is part of it.
The article goes into exactly why this happens!
That's exactly how you end up with mystery missing characters in something that's supposed to be evidence
> We see that that’s a quite a long line. Mail servers don’t like that
Why do mail server care about how long a line is? Why don't they just let the client reading the mail worry about wrapping the lines?
SMTP is a line–based protocol, including the part that transfers the message body
The server needs to parse the message headers, so it can't be an opaque blob. If the client uses IMAP, the server needs to fully parse the message. The only alternative is POP3, where the client downloads all messages as blobs and you can only read your email from one location, which made sense in the year 2000 but not now when everyone has several devices.
Hey, POP3 still makes sense. Having a local copy of your emails is useful.
If you want it to be the only copy and not sync with anything
POP3 is line–based too, anyway. Maybe you can rsync your maildir?
I just read it mainly in one place and through the web interface when I have to.
If your "in one place" reader is still open and downloading messages then there will be no messages to view in the web interface when you have to.
There will, because my client doesn't delete the messages from the server when it downloads them.
POP3 is more for reading and acting on your email in one place (taking notes, plan actions, discard and delete,…). No need to consume them on other devices as you’ve already extracted the important bits.
I use imap on my mobile device, but that’s mostly for recent emails until I get to my computer. Then it’s downloaded and deleted from the server.
But it's more akin to consuming a message queue. You have fetched it, it's gone.
Mails are (or used to be) processed line-by-line, typically using fixed-length buffers. This avoids dynamic memory allocation and having to write a streaming parser. RFC 821 finally limited the line length to at most 1000 bytes.
Given a mechanism for soft line breaks, breaking already at below 80 characters would increase compatibility with older mail software and be more convenient when listing the raw email in a terminal.
This is also why MIME Base64 typically inserts line breaks after 76 characters.
In early days, many/most people also read their email on terminals (or printers) with 80-column lines, so breaking lines at 72-ish was considered good email etiquette (to allow for later quoting prefix ">" without exceeding 80 characters).
One of the technical marvels of the day were mail and usenet clients that could properly render quoted text from infinite, never ending flame wars!
I don't think kids today realize how little memory we had when SMTP was designed.
For example, the PDP-11 (early 1970s), which was shared among dozens of concurrent users, had 512 kilobytes of RAM. The VAX-11 (late 1970s) might have as much as 2 megabytes.
Programmers were literally counting bytes to write programs.
"BITNET was a co-operative university computer network in the United States founded in 1981 by Ira Fuchs at the City University of New York (CUNY) and Greydon Freeman at Yale University."
https://en.wikipedia.org/wiki/BITNET
BITNET connected mainframes, had gateways to the Unix world and was still active in the 90s. And limited line lengths … some may remember SYSIN DD DATA … oh my goodness …
https://www.ibm.com/docs/en/zos/2.1.0?topic=execution-systsi...
This is how email work(ed) over smtp. When each command was sent it would get a '200'-class message (success) or 400/500-class message (failure). Sound familiar?
telnet smtp.mailserver.com 25
HELO
MAIL FROM: me@foo.com
RCPT TO: you@bar.com
DATA
blah blah blah
how's it going?
talk to you later!
.
QUIT
For anyone who wants to try this against a modern server:
This brings back some fun memories from the 1990s when this was exactly how we would send prank emails.
Yep! And also, if you included a blank line and then the headers for a new email in the bottom of your message, you could tell the server, hey, here comes another email for you to process!
If you were typing into a feedback form powered by something from Matt’s Script Archive, there was about a 95% chance you could trivially get it to send out multiple emails to other parties for every one email sent to the site’s owner.
That was nice part of 1990s - many systems allow for funny things ;)
I like how SMTP was at least honest in calling it the "receipt to" address and not the "sender" address.
Edit: wrong.
RCPT TO specifies the destination (recipient) address, the "sender" is what is written in MAIL FROM.
However what most mail programs show as sender and recipient is neither, they rather show the headers contained in the message.
Ah, sorry. You're right.
Back in 80s-90s it was common to use static buffers to simplify implementation - you allocate a fixed size buffer and reject a message if it has a line longer than the buffer size. SMTP RFC specifies 1000 symbols limit (including \r\n) but it's common to wrap around 87 symbols so it is easy to examine source (on a small screen).
The simplest reason: Mail servers have long had features which will send the mail client a substring of the text content without transferring the entire thing. Like the GMail inbox view, before you open any one message.
I suspect this is relevant because Quoted Printable was only a useful encoding for MIME types like text and HTML (the human readable email body), not binary (eg. Attachments, images, videos). Mail servers (if they want) can effectively treat the binary types as an opaque blob, while the text types can be read for more efficient transfer of message listings to the client.
As far as I can remember, most mail servers were fairly sane about that sort of thing, even back in the 90’s when this stuff was introduced. However, there were always these more or less motivated fears about some server somewhere running on some ancient IBM hardware using EBCDIC encoding and truncating everything to 72 characters because its model of the world was based on punched cards. So standards were written to handle all those bizarre systems. And I am sure that there is someone on HN who actually used one of those servers...
EBCDIC wasn't the problem, this was (part of) the problem:
https://www.ibm.com/docs/en/zos/2.1.0?topic=execution-systsi...
And BITNET …
> EBCDIC wasn't the problem
Wake up, everyone! Brand new sentence just dropped!
Thanks, I really expected a tale from the 70's, but did not see punch cards coming :)
The influence of 80 column punch cards remains pervasive.
RFC822 explicitly says it is for readability on systems with simple display software. Given that the protocol is from 1982 and systems back then had between 4 and 16kb RAM in total it might have made sense to give the lower end thin client systems of the day something preprocessed.
Also it is an easy way to stop a denial of service attack. If you let an infinite amount in that field. I can remotely overflow your system memory. The mail system can just error out and hang up on the person trying the attack instead of crashing out.
Surely you don't need the message to be broken up into lines just for that. Just read until a threshold is reached and then close the connection.
Keep in mind that in ye olden days, email was not a worldwide communication method. It was more typical for it to be an internal-only mail system, running on whatever legacy mainframe your org had, and working within whatever constraints that forced. So in the 90s when the internet began to expand, and email to external organizations became a bigger thing, you were just as concerned with compatibility with all those legacy terminal-based mail programs, which led to different choices when engineering the systems.
This is incorrect
Are you certain? Not OP, but a huge chunk of early RFCs was about how to let giant IBM systems talk to everyone else, specifying everything from character sets (nearly universally “7-bit ASCII”) to end of line/message characters. Otherwise, IBM would’ve tried to make EBCDIC the default for everything.
For instance, consider FTP’s text mode, which was primarily a way to accidentally corrupt your download when you forgot to type “bin” first, but was also handy for getting human readable files from one incompatible system to another.
I thought the article would be about the various meanings of operators like = == === .=. <== ==> <<== ==>> (==) => =~=
What is this, a Haskell for ants?
It has to be at least… three times bigger than this
My fist association was brainf..k (*.bf) programming language
This ended up being way more interesting
I'm just wondering why this problem shows up now. Why do lots of people suddenly post their old emails with a defective QP decoder?
> For some reason or other, people have been posting a lot of excerpts from old emails on Twitter over the last few days.
On the risk of having missed the latest meme or social media drama, but does anyone know what this "some reason or other" is?
Edit: Question answered.
Presumably the Epstein files, but I'm not on twitter so not sure
you presume cor=rect
https://www.jmail.world/thread/EFTA02512824?view=person
Huh, Noam Chomsky, nice one!
Ooh, that reason. Sorry for having been dense. Thanks!
Jeff Epstein? The New York financier?
the DOJ published another bunch of Epstein emails
I wrote my own email archiving software. The hardest part was dealing with all the weird edge cases in my 20+ year collection of .eml files. For being so simple conceptually, email is surprisingly complicated.
I wrote a console-based mail client, which was 25% C++ and 75% Lua for defining the UI and the processing.
It never got too popular, but I had users for a few years and I can honestly say MIME was the bane of my life for most of those years.
> So what’s happened here? Well, whoever collected these emails first converted from CRLF (i.e., “Windows” line ending coding) to “NL” (i.e., “Unix” line ending coding). This is pretty normal if you want to deal with email. But you then have one byte fewer:
I think there is a second possible conclusion, which is that the transformation happened historically. Everyone assumes these emails are an exact dump from Gmail, but isn't it possible that Epstein was syncing emails from Gmail to a third party mail server?
Since the Stackoverflow post details the exact situation in 2011, I think we should be open to the idea that we're seeing data collected from a secondary mail server, not Gmail directly.
Do we have anything to discount this?
(If I'm not mistaken, I think you can also see the "=" issue simply by applying the Quoted-Printable encoding twice, not just by mishandling the line-endings, which also makes me think two mail servers. It also explains why the "=" symbol is retained.)
In one of the email PDFs I saw an XML plist with some metadata that looked like it was from Apple's Mail.app, so these might be extracted from whatever internal format that uses.
Yeah, I wouldn't bet on this being a single bad Gmail export; it smells much more like the accumulated scars of multiple mail systems doing "helpful" things to the same messages over time
This seems like the most likely reason to me!
What's funny is that the failure mode here is so quietly destructive
(The title of the blog reminded me the late Bob Pease [1] who had the signature, "What's all this XXX stuff, anyhow?" [2] where XXX might be "noise gain", "capacitor leakage"…)
[1] https://en.wikipedia.org/wiki/Bob_Pease
[2] https://www.qsl.net/n9zia/pease/index.html
Fun how the archive.today article near the top has this exact issue
https://pastes.io/correspond
https://news.ycombinator.com/item?id=46843805
https://web.archive.org/web/20260203094902/https://lars.inge...
Did the site get the HN kiss of death?
CLRF vs LF strikes again. Partly at least.
I wonder why even have a max line length limit in the first place? I.e. is this for a technical reason or just display related?
Wait, now we have to deal with Carriage Line Return Feeds too?
I wonder if the person who had the idea of virtualizing the typewriter carriage knew how much trouble they would cause over time.
Yeah, and using two bytes for a single line termination (or separation or whatever)? Why make things more complicated and take more space at the same time?
Remember that back in the mists of time, computers used typewriter-esque machines for user interaction and text output. You had to send a CR followed by an LF to go to the next line on the physical device. Storing both characters in the file meant the OS didn't need to insert any additional characters when printing. Having two separate characters let you do tricks like overstriking (just send CR, no LF)
True, but I don’t think there was a common reason to ever send a linefeed without going back to the beginning. Were people printing lots of vertical pipe characters at column 70 or something?
It would’ve been far less messy to make printers process linefeed like \n acts today, and omit the redundant CR. Then you could still use CR for those overstrike purposes but have a 1-byte universal newline character, which we almost finally have today now that Windows mostly stopped resisting the inevitable.
I haven't seen them other than in the submission - but if the length matches up it may be that they were processed from raw email, the RFC defines a length to wrap at.
Edit: yes I think that's most likely what it is (and it's SHOULD 78ch; MUST 998ch) - I was forgetting that it also specifies the CRLF usage, it's not (necessarily) related to Windows at all here as described in TFA.
Here it is in my 'notmuch-more' email lib: https://github.com/OJFord/amail/blob/8904c91de6dfb5cba2b279f...
> it's not (necessarily) related to Windows at all here as described in TFA.
The article doesn't claim that it's Windows related. The article is very clear in explaining that the spec requires =CRLF (3 characters), then mentions (in passing) that CRLF is the typical line ending on Windows, then speculates that someone replaced the two characters CRLF with a one character new line, as on Unix or other OSs.
Ok yeah I may have misinterpreted that bit in the article. It would be a totally reasonable assumption if you didn't happen to know that about email though, it wasn't a judgement regardless.
I am just wondering how it is good idea for a sever to insert some characters into user's input. If a collegue were to propose this, i d laugh in his face
It's just sp hacky i cant belive it's a real life's solution
“Insert characters”?
Consider converting the original text (maintaining the author’s original line wrapping and indentation) to base64. Has anything been “inserted” into the text? I would suggest not. It has been encoded.
Now consider an encoding that leaves most of the text readable, translates some things based on a line length limit, and some other things based on transport limitations (e.g. passing through 7-bit systems.) As long as one follows the correct decoding rules, the original will remain intact - nothing “inserted.” The problem is someone just knowledgeable enough to be aware that email is human readable but not aware of the proper decoding has attempted to “clean up” the email for sharing.
Okey it does sound better from this POV. Still wierd as its a Client/UI concern, not something a server is supposed to do; whats next,adding "bold" tags on the title? Lol
SMTP is a line-oriented protocol. The server processes one line at a time, and needs to understand headers.
Infinite line length = infinite buffer. Even worse, QP is 7-bit (because SMTP started out ASCII only), so characters >127 get encoded as three bytes (equal, then two hex digits), so a 500-character non-ASCII UTF8 line is 1500 bytes.
It all made sense at the time. Not so much these days when 7-bit pipes only exist because they always have.
When you post a comment on HN, the server inserts HTML tags into your input. Isn't that essentially the same thing?
No, because there is a clear separation between the content and the envelop. You wouldnt expect the post office to open your physical letters and write routing instructions to the postmen for delivery
But I agree with sibling comment: it makes more sense when its called "encoding" instead of "inserting chars into original stream"
> You wouldnt expect the post office to open your physical letters and write routing instructions to the postmen for delivery
Digital communication is based on the postmen reading, transcribing and copying your letters. There is a reason why digital communication is treated differently then letters by the law and why the legally mandated secrecy for letters doesn't apply to emails.
It's called escaping, and almost every protocol has it. HN must convert the & symbol to & for displaying in HTML. Many wire protocols like SATA or Ethernet must insert a 1 after a certain number of consecutive 0s to maintain electrical balance. Don't remember which ones — don't quote me that it's SATA and Ethernet.
Protocols that literally insert a bit are HDLC / PPP / CAN and they insert a 0 after a few 1s
Just wait until you learn what mess UTF-8 will turn your characters into. ;)
I love how HN always floats up the answers to questions that were in my mind, without occupying my mind.
I, too, was reading about the new Epstein files, wondering what text artifact was causing things to look like that.
Same here. I did notice what I think was an actual error on someone's part, there was a chart in the files comparing black to white IQ distributions, and well, just look at it:
https://nitter.net/AFpost/status/2017415163763429779?s=201
Something clearly went wrong in the process.
Me too. I first assumced it was an OCR error, then remembered they were emails and wouldn't need to go through OCR. Then I thought that the US Government is exactly the kind of place to print out millions of emails only to scan them back in again.
I'm glad to know the real reason!
My main takeaway from this article, is that I want to know what happened to the modified pigs with non-cloven hoofs
Great. Can't wait for equal signs to be the next (((whatever this is))). Maybe it's a secret code. j/k
On a side note: There are actually products marketed as kosher bacon (it's usually beef or turkey). And secular Jews frequently make jokes like this about our kosher bros who aren't allowed to eat the real stuff for some dumb reason like it has too many toes.
But... pigs _do_ have cloven hooves! The issue is that they're not ruminant.
That said, there is a _possibly_ kosher pig: https://en.wikipedia.org/wiki/Babirusa#Relationship_with_hum...
"It’s a fascinating case of 'Abstraction Leak'.
We’ve become so accustomed to modern libraries handling encoding transparently that when raw data surfaces (like in these dumps), we often lack the 'Digital Archeology' skills to recognize basic Quoted-Printable.
These artifacts (=20, =3D) are effectively fossils of the transport layer. It’s a stark reminder that underneath our modern AI/React/JSON world, the internet is still largely held together by 7-bit ASCII constraints and protocols from the 1980s.
TLDR "=\r\n" was converted to "=\n"
Author seems to think Unix uses a character called "NL" instead of "LF"...
Unicode labels U+000A as all of "LINE FEED (LF)", "new line (NL)" and "end of line (EOL)". I'm guessing different names were imported from slightly different character sets, although I understand the all-uppercase name to be the main/official one.
https://www.unicode.org/charts/PDF/U0000.pdf
NL, or New Line, is a character in some character sets, like old mainframe computers. No need to be snarky just because he mistyped or uses a different name for something.
I am more surprised by the description of “rock döts”. A Norwegian certainly knows that ASCII is not enough for all our alphabetical needs.
https://en.wikipedia.org/wiki/Metal_umlaut
The writer presumably knows that umlauts and other non-ascii characters are functional in many languages. "rock döts" is poking fun at the trend in a certain tranche of anglophone rock/metal to use them in a purely aesthetic way in band names etc.
No, the article is quite explicit that that isn't what happened.
Could be worsened by inaccurate optical character recognition in some cases.
Back in those days optical scanners were still used.
People posting Excel formulae?
Rock dots? You mean diacritics? Yeah someone invented them: the ancient Greeks, idiöt.
It's not the character, its the way / context in which it's used
https://en.wikipedia.org/wiki/Metal_umlaut
I know what he was referring to. But the use case is obviously languages other than English, not the Motörhead fan club newsletter.
Some combination of people misunderstood some other people's joke, not totally clear which and which.
Yeah, that dude oughta read books and learn about computers, too.
And live in a country where they use these in their alphabets.