update parsing for manually configured proxy on Windows

[wfurt]

Fixes#38670
Fixes#91045

The current behavior was intruded by dotnet/corefx#40082 to support multiple proxies.
The logic for parsing PAC files was also applied to the manual proxy setting.
But as the issues above noticed that does not quite reflect reality.
The existing code assumes that ion the proxy location starts with http:// it would apply only to HTTP (and not HTTPS traffic. In reality one can put http://x.y.x/ string to the proxy location and all browsers would apply it to both http & https.
If the string is prefixed with http= it is also respected e.g. http=http://x.y.x/ would apply only for HTTP and not HTTPS traffic.

To limit regressions I split the logic and tests and I left PAC processing ASIS and I made small change to relax the restrictions if it comes from the manual settings location.

[MihaZupan]

LGTM.
I take it automatic proxy discovery does want something like http=https://proxy.secure.com to only apply to https traffic? Or was that what you meant by trying to limit regressions and not modifying that part?

[wfurt]

The http= and https= is somewhat easy. The prefix specifies what traffic kind (e.g. http or https) the settings applies to and what URI to go to e.g. you can have http=:http://1.2.3.4:3128, https=http://1.2.3.4:3128 and that would apply proxy to both http & https using some good old plain HTTP proxy (TLS via CONNECT). Things get more shady when the prefix is omitted. The old code assumed that the if the proxy destination is http://x.x.x..x it would apply only to plain HTTP. (I don't know if Windows support secure proxies but I could imagine something like http=https://x.y.z/ but the syntax without)
But that assumption does not work for the manual proxy configuration. I did not do extensive research on the PAC options but too limit risk I left the existing code as is e.g. unless I missed something the PAC should be processed exactly as it is today.

[wfurt]

Fixes#38670
Fixes#91045

The current behavior was intruded by dotnet/corefx#40082 to support multiple proxies.
The logic for parsing PAC files was also applied to the manual proxy setting.
But as the issues above noticed that does not quite reflect reality.
The existing code assumes that ion the proxy location starts with http:// it would apply only to HTTP (and not HTTPS traffic. In reality one can put http://x.y.x/ string to the proxy location and all browsers would apply it to both http & https.
If the string is prefixed with http= it is also respected e.g. http=http://x.y.x/ would apply only for HTTP and not HTTPS traffic.

To limit regressions I split the logic and tests and I left PAC processing ASIS and I made small change to relax the restrictions if it comes from the manual settings location.

[MihaZupan]

LGTM.
I take it automatic proxy discovery does want something like http=https://proxy.secure.com to only apply to https traffic? Or was that what you meant by trying to limit regressions and not modifying that part?

[wfurt]

The http= and https= is somewhat easy. The prefix specifies what traffic kind (e.g. http or https) the settings applies to and what URI to go to e.g. you can have http=:http://1.2.3.4:3128, https=http://1.2.3.4:3128 and that would apply proxy to both http & https using some good old plain HTTP proxy (TLS via CONNECT). Things get more shady when the prefix is omitted. The old code assumed that the if the proxy destination is http://x.x.x..x it would apply only to plain HTTP. (I don't know if Windows support secure proxies but I could imagine something like http=https://x.y.z/ but the syntax without)
But that assumption does not work for the manual proxy configuration. I did not do extensive research on the PAC options but too limit risk I left the existing code as is e.g. unless I missed something the PAC should be processed exactly as it is today.

[wfurt]

Fixes#38670
Fixes#91045

The current behavior was intruded by dotnet/corefx#40082 to support multiple proxies.
The logic for parsing PAC files was also applied to the manual proxy setting.
But as the issues above noticed that does not quite reflect reality.
The existing code assumes that ion the proxy location starts with http:// it would apply only to HTTP (and not HTTPS traffic. In reality one can put http://x.y.x/ string to the proxy location and all browsers would apply it to both http & https.
If the string is prefixed with http= it is also respected e.g. http=http://x.y.x/ would apply only for HTTP and not HTTPS traffic.

To limit regressions I split the logic and tests and I left PAC processing ASIS and I made small change to relax the restrictions if it comes from the manual settings location.

[MihaZupan]

LGTM.
I take it automatic proxy discovery does want something like http=https://proxy.secure.com to only apply to https traffic? Or was that what you meant by trying to limit regressions and not modifying that part?

[wfurt]

The http= and https= is somewhat easy. The prefix specifies what traffic kind (e.g. http or https) the settings applies to and what URI to go to e.g. you can have http=:http://1.2.3.4:3128, https=http://1.2.3.4:3128 and that would apply proxy to both http & https using some good old plain HTTP proxy (TLS via CONNECT). Things get more shady when the prefix is omitted. The old code assumed that the if the proxy destination is http://x.x.x..x it would apply only to plain HTTP. (I don't know if Windows support secure proxies but I could imagine something like http=https://x.y.z/ but the syntax without)
But that assumption does not work for the manual proxy configuration. I did not do extensive research on the PAC options but too limit risk I left the existing code as is e.g. unless I missed something the PAC should be processed exactly as it is today.

[wfurt]

Fixes#38670
Fixes#91045

The current behavior was intruded by dotnet/corefx#40082 to support multiple proxies.
The logic for parsing PAC files was also applied to the manual proxy setting.
But as the issues above noticed that does not quite reflect reality.
The existing code assumes that ion the proxy location starts with http:// it would apply only to HTTP (and not HTTPS traffic. In reality one can put http://x.y.x/ string to the proxy location and all browsers would apply it to both http & https.
If the string is prefixed with http= it is also respected e.g. http=http://x.y.x/ would apply only for HTTP and not HTTPS traffic.

To limit regressions I split the logic and tests and I left PAC processing ASIS and I made small change to relax the restrictions if it comes from the manual settings location.

[MihaZupan]

LGTM.
I take it automatic proxy discovery does want something like http=https://proxy.secure.com to only apply to https traffic? Or was that what you meant by trying to limit regressions and not modifying that part?

[wfurt]

The http= and https= is somewhat easy. The prefix specifies what traffic kind (e.g. http or https) the settings applies to and what URI to go to e.g. you can have http=:http://1.2.3.4:3128, https=http://1.2.3.4:3128 and that would apply proxy to both http & https using some good old plain HTTP proxy (TLS via CONNECT). Things get more shady when the prefix is omitted. The old code assumed that the if the proxy destination is http://x.x.x..x it would apply only to plain HTTP. (I don't know if Windows support secure proxies but I could imagine something like http=https://x.y.z/ but the syntax without)
But that assumption does not work for the manual proxy configuration. I did not do extensive research on the PAC options but too limit risk I left the existing code as is e.g. unless I missed something the PAC should be processed exactly as it is today.

[wfurt]

Fixes#38670
Fixes#91045

The current behavior was intruded by dotnet/corefx#40082 to support multiple proxies.
The logic for parsing PAC files was also applied to the manual proxy setting.
But as the issues above noticed that does not quite reflect reality.
The existing code assumes that ion the proxy location starts with http:// it would apply only to HTTP (and not HTTPS traffic. In reality one can put http://x.y.x/ string to the proxy location and all browsers would apply it to both http & https.
If the string is prefixed with http= it is also respected e.g. http=http://x.y.x/ would apply only for HTTP and not HTTPS traffic.

To limit regressions I split the logic and tests and I left PAC processing ASIS and I made small change to relax the restrictions if it comes from the manual settings location.

[MihaZupan]

LGTM.
I take it automatic proxy discovery does want something like http=https://proxy.secure.com to only apply to https traffic? Or was that what you meant by trying to limit regressions and not modifying that part?

[wfurt]

The http= and https= is somewhat easy. The prefix specifies what traffic kind (e.g. http or https) the settings applies to and what URI to go to e.g. you can have http=:http://1.2.3.4:3128, https=http://1.2.3.4:3128 and that would apply proxy to both http & https using some good old plain HTTP proxy (TLS via CONNECT). Things get more shady when the prefix is omitted. The old code assumed that the if the proxy destination is http://x.x.x..x it would apply only to plain HTTP. (I don't know if Windows support secure proxies but I could imagine something like http=https://x.y.z/ but the syntax without)
But that assumption does not work for the manual proxy configuration. I did not do extensive research on the PAC options but too limit risk I left the existing code as is e.g. unless I missed something the PAC should be processed exactly as it is today.

[wfurt]

Fixes#38670
Fixes#91045

The current behavior was intruded by dotnet/corefx#40082 to support multiple proxies.
The logic for parsing PAC files was also applied to the manual proxy setting.
But as the issues above noticed that does not quite reflect reality.
The existing code assumes that ion the proxy location starts with http:// it would apply only to HTTP (and not HTTPS traffic. In reality one can put http://x.y.x/ string to the proxy location and all browsers would apply it to both http & https.
If the string is prefixed with http= it is also respected e.g. http=http://x.y.x/ would apply only for HTTP and not HTTPS traffic.

To limit regressions I split the logic and tests and I left PAC processing ASIS and I made small change to relax the restrictions if it comes from the manual settings location.

[MihaZupan]

LGTM.
I take it automatic proxy discovery does want something like http=https://proxy.secure.com to only apply to https traffic? Or was that what you meant by trying to limit regressions and not modifying that part?

[wfurt]

The http= and https= is somewhat easy. The prefix specifies what traffic kind (e.g. http or https) the settings applies to and what URI to go to e.g. you can have http=:http://1.2.3.4:3128, https=http://1.2.3.4:3128 and that would apply proxy to both http & https using some good old plain HTTP proxy (TLS via CONNECT). Things get more shady when the prefix is omitted. The old code assumed that the if the proxy destination is http://x.x.x..x it would apply only to plain HTTP. (I don't know if Windows support secure proxies but I could imagine something like http=https://x.y.z/ but the syntax without)
But that assumption does not work for the manual proxy configuration. I did not do extensive research on the PAC options but too limit risk I left the existing code as is e.g. unless I missed something the PAC should be processed exactly as it is today.

[wfurt]

Fixes#38670
Fixes#91045

The current behavior was intruded by dotnet/corefx#40082 to support multiple proxies.
The logic for parsing PAC files was also applied to the manual proxy setting.
But as the issues above noticed that does not quite reflect reality.
The existing code assumes that ion the proxy location starts with http:// it would apply only to HTTP (and not HTTPS traffic. In reality one can put http://x.y.x/ string to the proxy location and all browsers would apply it to both http & https.
If the string is prefixed with http= it is also respected e.g. http=http://x.y.x/ would apply only for HTTP and not HTTPS traffic.

To limit regressions I split the logic and tests and I left PAC processing ASIS and I made small change to relax the restrictions if it comes from the manual settings location.

[MihaZupan]

LGTM.
I take it automatic proxy discovery does want something like http=https://proxy.secure.com to only apply to https traffic? Or was that what you meant by trying to limit regressions and not modifying that part?

[wfurt]

The http= and https= is somewhat easy. The prefix specifies what traffic kind (e.g. http or https) the settings applies to and what URI to go to e.g. you can have http=:http://1.2.3.4:3128, https=http://1.2.3.4:3128 and that would apply proxy to both http & https using some good old plain HTTP proxy (TLS via CONNECT). Things get more shady when the prefix is omitted. The old code assumed that the if the proxy destination is http://x.x.x..x it would apply only to plain HTTP. (I don't know if Windows support secure proxies but I could imagine something like http=https://x.y.z/ but the syntax without)
But that assumption does not work for the manual proxy configuration. I did not do extensive research on the PAC options but too limit risk I left the existing code as is e.g. unless I missed something the PAC should be processed exactly as it is today.

[wfurt]

Fixes#38670
Fixes#91045

The current behavior was intruded by dotnet/corefx#40082 to support multiple proxies.
The logic for parsing PAC files was also applied to the manual proxy setting.
But as the issues above noticed that does not quite reflect reality.
The existing code assumes that ion the proxy location starts with http:// it would apply only to HTTP (and not HTTPS traffic. In reality one can put http://x.y.x/ string to the proxy location and all browsers would apply it to both http & https.
If the string is prefixed with http= it is also respected e.g. http=http://x.y.x/ would apply only for HTTP and not HTTPS traffic.

To limit regressions I split the logic and tests and I left PAC processing ASIS and I made small change to relax the restrictions if it comes from the manual settings location.

[MihaZupan]

LGTM.
I take it automatic proxy discovery does want something like http=https://proxy.secure.com to only apply to https traffic? Or was that what you meant by trying to limit regressions and not modifying that part?

[wfurt]

The http= and https= is somewhat easy. The prefix specifies what traffic kind (e.g. http or https) the settings applies to and what URI to go to e.g. you can have http=:http://1.2.3.4:3128, https=http://1.2.3.4:3128 and that would apply proxy to both http & https using some good old plain HTTP proxy (TLS via CONNECT). Things get more shady when the prefix is omitted. The old code assumed that the if the proxy destination is http://x.x.x..x it would apply only to plain HTTP. (I don't know if Windows support secure proxies but I could imagine something like http=https://x.y.z/ but the syntax without)
But that assumption does not work for the manual proxy configuration. I did not do extensive research on the PAC options but too limit risk I left the existing code as is e.g. unless I missed something the PAC should be processed exactly as it is today.