Re: Javamail address parsing (again).
Dain Sundstrom wrote: On Jan 12, 2006, at 3:24 PM, Rick McGuire wrote: Dain Sundstrom wrote: On Jan 11, 2006, at 1:17 PM, Bruce Snyder wrote: Is it possible to look at the Sun implementation's source code to distinguish enforced vs. ignored rules? That would make the code not clean room. I propose we ask Sun for a formal definition of the parser for this class, and in a parallel track make an effort to try to match their bugs. The code from the second track doesn't have to be perfect, but just good enough. We simply let our users know that our goal with the "implement.sun.javamail.bugs=true" code is to emulate the sun bugs, and if they find something that produces different results for the same text, we consider it a bug. I'm becoming less and less convinced this is a good idea. So far, I've found many, many sun bugs in this code where they produce results that are in conflict with with RFC822. The API documentation refers relaxed parsing rules, which says to me there are addresses that would not be valid under RFC822, but javamail will accept them based on the type of parsing requested. I can accept that. However, the great majority of the problems I've found have been involved with internet addresses that RFC822 says ARE valid, but the javamail code does not handle them properly. And there are a few situations where it appears the authors just chose to punt and say "yeah, whatever". It appears that the solution is to write hacked code that mostly, sorta, kinda does what it claims to do, or write a good parser, then triple the size of the code trying to get all of the Sun bugs to work properly. Working strictly from the RFC822 spec, I had a fairly nice parser written that gave very good RFC822 compliance, but things turned nightmarish when I discovered the sorts of Sun behaviors I had to insert back in. I think I've completely rewritten this code about 5 times now, and am getting pretty close to the Sun "relaxed rules". Inserting some of the real bugs back in to the parsing might pose similar problems. It really appears that this code somewhat "lost it's way somewhere". It's serving two purposes that are really at odds with each other. The first purpose, is to parse any internet address that might appear in a received message. For that purpose, the code needs to accept any valid internet address as defined by RFC822. The Sun code does not currently do that, and making the new version "bug compatible" would also not achieve that. The other purpose of the InternetAddress parser is to process email addresses entered into applications and perform some validation on the addresses. This is where the "relaxed rules" come in to play, and basically allows internet addresses that are not strictly RFC822 compatible to pass. Now for those, I'm relatively comfortable that this can be made compatible. It is very difficult though, when the requirement becomes one of being both more and less restrictive at the same time, with no good definition of the what rules are being used. Ok, how about we say, in "sun bug mode" we will parse all addresses that are valid RFC822 address or are sucessfull parsed by sun's javamail implementation? This means that valid RFC822 addresses that sun's implementation rejects will be accepted by ours. We would further would consider it a bug to reject addresses accepted by sun's implementation when in "sun bug mode". That sounds like a more reasonable goal. -dain
Re: Javamail address parsing (again).
On Jan 12, 2006, at 3:24 PM, Rick McGuire wrote: Dain Sundstrom wrote: On Jan 11, 2006, at 1:17 PM, Bruce Snyder wrote: Is it possible to look at the Sun implementation's source code to distinguish enforced vs. ignored rules? That would make the code not clean room. I propose we ask Sun for a formal definition of the parser for this class, and in a parallel track make an effort to try to match their bugs. The code from the second track doesn't have to be perfect, but just good enough. We simply let our users know that our goal with the "implement.sun.javamail.bugs=true" code is to emulate the sun bugs, and if they find something that produces different results for the same text, we consider it a bug. I'm becoming less and less convinced this is a good idea. So far, I've found many, many sun bugs in this code where they produce results that are in conflict with with RFC822. The API documentation refers relaxed parsing rules, which says to me there are addresses that would not be valid under RFC822, but javamail will accept them based on the type of parsing requested. I can accept that. However, the great majority of the problems I've found have been involved with internet addresses that RFC822 says ARE valid, but the javamail code does not handle them properly. And there are a few situations where it appears the authors just chose to punt and say "yeah, whatever". It appears that the solution is to write hacked code that mostly, sorta, kinda does what it claims to do, or write a good parser, then triple the size of the code trying to get all of the Sun bugs to work properly. Working strictly from the RFC822 spec, I had a fairly nice parser written that gave very good RFC822 compliance, but things turned nightmarish when I discovered the sorts of Sun behaviors I had to insert back in. I think I've completely rewritten this code about 5 times now, and am getting pretty close to the Sun "relaxed rules". Inserting some of the real bugs back in to the parsing might pose similar problems. It really appears that this code somewhat "lost it's way somewhere". It's serving two purposes that are really at odds with each other. The first purpose, is to parse any internet address that might appear in a received message. For that purpose, the code needs to accept any valid internet address as defined by RFC822. The Sun code does not currently do that, and making the new version "bug compatible" would also not achieve that. The other purpose of the InternetAddress parser is to process email addresses entered into applications and perform some validation on the addresses. This is where the "relaxed rules" come in to play, and basically allows internet addresses that are not strictly RFC822 compatible to pass. Now for those, I'm relatively comfortable that this can be made compatible. It is very difficult though, when the requirement becomes one of being both more and less restrictive at the same time, with no good definition of the what rules are being used. Ok, how about we say, in "sun bug mode" we will parse all addresses that are valid RFC822 address or are sucessfull parsed by sun's javamail implementation? This means that valid RFC822 addresses that sun's implementation rejects will be accepted by ours. We would further would consider it a bug to reject addresses accepted by sun's implementation when in "sun bug mode". -dain
Re: Javamail address parsing (again).
Dain Sundstrom wrote: On Jan 11, 2006, at 1:17 PM, Bruce Snyder wrote: Is it possible to look at the Sun implementation's source code to distinguish enforced vs. ignored rules? That would make the code not clean room. I propose we ask Sun for a formal definition of the parser for this class, and in a parallel track make an effort to try to match their bugs. The code from the second track doesn't have to be perfect, but just good enough. We simply let our users know that our goal with the "implement.sun.javamail.bugs=true" code is to emulate the sun bugs, and if they find something that produces different results for the same text, we consider it a bug. I'm becoming less and less convinced this is a good idea. So far, I've found many, many sun bugs in this code where they produce results that are in conflict with with RFC822. The API documentation refers relaxed parsing rules, which says to me there are addresses that would not be valid under RFC822, but javamail will accept them based on the type of parsing requested. I can accept that. However, the great majority of the problems I've found have been involved with internet addresses that RFC822 says ARE valid, but the javamail code does not handle them properly. And there are a few situations where it appears the authors just chose to punt and say "yeah, whatever". It appears that the solution is to write hacked code that mostly, sorta, kinda does what it claims to do, or write a good parser, then triple the size of the code trying to get all of the Sun bugs to work properly. Working strictly from the RFC822 spec, I had a fairly nice parser written that gave very good RFC822 compliance, but things turned nightmarish when I discovered the sorts of Sun behaviors I had to insert back in. I think I've completely rewritten this code about 5 times now, and am getting pretty close to the Sun "relaxed rules". Inserting some of the real bugs back in to the parsing might pose similar problems. It really appears that this code somewhat "lost it's way somewhere". It's serving two purposes that are really at odds with each other. The first purpose, is to parse any internet address that might appear in a received message. For that purpose, the code needs to accept any valid internet address as defined by RFC822. The Sun code does not currently do that, and making the new version "bug compatible" would also not achieve that. The other purpose of the InternetAddress parser is to process email addresses entered into applications and perform some validation on the addresses. This is where the "relaxed rules" come in to play, and basically allows internet addresses that are not strictly RFC822 compatible to pass. Now for those, I'm relatively comfortable that this can be made compatible. It is very difficult though, when the requirement becomes one of being both more and less restrictive at the same time, with no good definition of the what rules are being used. -dain
Re: Javamail address parsing (again).
On Jan 11, 2006, at 1:17 PM, Bruce Snyder wrote: Is it possible to look at the Sun implementation's source code to distinguish enforced vs. ignored rules? That would make the code not clean room. I propose we ask Sun for a formal definition of the parser for this class, and in a parallel track make an effort to try to match their bugs. The code from the second track doesn't have to be perfect, but just good enough. We simply let our users know that our goal with the "implement.sun.javamail.bugs=true" code is to emulate the sun bugs, and if they find something that produces different results for the same text, we consider it a bug. -dain
Re: Javamail address parsing (again).
On 1/11/06, Rick McGuire <[EMAIL PROTECTED]> wrote: > This is starting to drive me nuts. Writing an address parsing method > that conforms to RFC822 is fairly easy. Writing one that conforms to > the javamail spec seems to be a hopeless task. This is the complete API > spec for the InternetAddress.parseHeader() method: > > Parse the given sequence of addresses into InternetAddress objects. > If |strict| is false, the full syntax rules for individual addresses > are not enforced. If |strict| is true, many (but not all) of the > RFC822 syntax rules are enforced. > > To better support the range of "invalid" addresses seen in real > messages, this method enforces fewer syntax rules than the |parse| > method when the strict flag is false and enforces more rules when > the strict flag is true. If the strict flag is false and the parse > is successful in separating out an email address or addresses, the > syntax of the addresses themselves is not checked. > > There is absolutely no definition I can find of: > > * What syntax rules are not enforced if strict is false. > * What syntax rules are not enforeced if strict is true. > * What is the difference in syntax rule enforcement between > parseHeader() and parse(). parse() seems to a rule set that lies > between parseHeader() with strict false and parseHeader with > strict true. > * What does it mean to be "successful in separating out an email > address or addresses" without checking the syntax? How do you > recognize it as an email address without having syntax rules? > > There don't appear to be any other sources of information available out > there that further define this behavior. I've been running lots of > little test cases against the Sun version to try to figure out the > rules, and frankly, the results have been pretty random. The Sun > version both allows forms that RFC822 says is invalid and rejects forms > that RFC822 explicitly says are valid (which does not sound like a > relaxed rule to me). Rather tough to distinguish between bugs and > intentional behavior. > > Any suggestions on additional information sources on this or suggestions > on how to decide which behaviors to support? Is it possible to look at the Sun implementation's source code to distinguish enforced vs. ignored rules? Bruce -- perl -e 'print unpack("u30","D0G)[EMAIL PROTECTED]&5R\"F)R=6-E+G-N>61Ehttp://geronimo.apache.org/) Castor (http://castor.org/)
Javamail address parsing (again).
This is starting to drive me nuts. Writing an address parsing method that conforms to RFC822 is fairly easy. Writing one that conforms to the javamail spec seems to be a hopeless task. This is the complete API spec for the InternetAddress.parseHeader() method: Parse the given sequence of addresses into InternetAddress objects. If |strict| is false, the full syntax rules for individual addresses are not enforced. If |strict| is true, many (but not all) of the RFC822 syntax rules are enforced. To better support the range of "invalid" addresses seen in real messages, this method enforces fewer syntax rules than the |parse| method when the strict flag is false and enforces more rules when the strict flag is true. If the strict flag is false and the parse is successful in separating out an email address or addresses, the syntax of the addresses themselves is not checked. There is absolutely no definition I can find of: * What syntax rules are not enforced if strict is false. * What syntax rules are not enforeced if strict is true. * What is the difference in syntax rule enforcement between parseHeader() and parse(). parse() seems to a rule set that lies between parseHeader() with strict false and parseHeader with strict true. * What does it mean to be "successful in separating out an email address or addresses" without checking the syntax? How do you recognize it as an email address without having syntax rules? There don't appear to be any other sources of information available out there that further define this behavior. I've been running lots of little test cases against the Sun version to try to figure out the rules, and frankly, the results have been pretty random. The Sun version both allows forms that RFC822 says is invalid and rejects forms that RFC822 explicitly says are valid (which does not sound like a relaxed rule to me). Rather tough to distinguish between bugs and intentional behavior. Any suggestions on additional information sources on this or suggestions on how to decide which behaviors to support? Rick