Re: Insufficiencies in JEP: 400: UTF-8 by Default
Hi Anthony, A draft of updates to the Process API is in the works and covers improving the ease of use and providing Readers and Writers. Note that if process output is redirected to a file, it does not interpose on the byte streams and is not in a position to affect the character set used by the child process. Regards, Roger On 3/30/21 1:03 PM, Anthony Vanelverdinghe wrote: Hi Alan As Marco mentioned, another use case is sub-process stdin/stdout/stderr. In my particular instance, I'm starting a Process which has its output redirected to a file. It uses the platform's default encoding for writing to stdout. So when I want to read its output from the file at some later point, I need to supply that encoding to the Files API. One way to accommodate this use case, is a method which allows to retrieve the platform's default encoding, for example a method `platformEncoding` in Charset or Process, or the `Console::charset` method you mentioned. Another option would be to enhance the Process API, by adding methods to Process which return appropriate Readers/Writers & adding methods of the form `redirectX(File file, Charset encoding)` to ProcessBuilder. But this seems like a lot of additional API surface, just to avoid surfacing the platform's default encoding itself. So I think the JEP should specify how it'll address use cases w.r.t. the Process API, shouldn't it? Kind regards, Anthony On Sunday, March 14, 2021 13:01 CET, Alan Bateman wrote: On 14/03/2021 11:00, Marco wrote: : IMO Charset should provide standardized getters for the OS charset and the console charset. The latter being different has been a long standing issue on Windows where the codepage differs between its CLI and regular environments. OpenJDK has the necessary data already available in its custom system properties. The console charset is currently hidden behind PrintStream not exposing the underlying OSWriter and not offering getEncoding() itself. The OS charset would be hidden in the future by Charset.getDefaultCharset()'s specification change in JEP 400. The intention that there will be little or no impact to the console streams. This means that java.io.Console reader/writer methods should continue to return a Reader/PrintWriter that uses the platform encoding (or code page is on Windows). Same thing for the System.out/System.err print streams. We need to make this clearer in the JEP. There has been discussion on this mailing list about adding a Console::charset method but it didn't come to a consensus. Naoto Sato and I have been chatting about it again recently as there may be a need to add an API in advance of proposing to target the JEP. One case that we are still mulling over is code that creates an InputStreamReader on System.in without specifying the charset. This might be older code that pre-dates java.io.Console or maybe code that wasn't tested on a wide range or platforms. Options range from a spec change to doing nothing (the latter meaning running with "COMPACT" or migrating the code to use the 2-arg constructor as the default charset is not the right choice). -Alan
Re: Insufficiencies in JEP: 400: UTF-8 by Default
Hi Alan As Marco mentioned, another use case is sub-process stdin/stdout/stderr. In my particular instance, I'm starting a Process which has its output redirected to a file. It uses the platform's default encoding for writing to stdout. So when I want to read its output from the file at some later point, I need to supply that encoding to the Files API. One way to accommodate this use case, is a method which allows to retrieve the platform's default encoding, for example a method `platformEncoding` in Charset or Process, or the `Console::charset` method you mentioned. Another option would be to enhance the Process API, by adding methods to Process which return appropriate Readers/Writers & adding methods of the form `redirectX(File file, Charset encoding)` to ProcessBuilder. But this seems like a lot of additional API surface, just to avoid surfacing the platform's default encoding itself. So I think the JEP should specify how it'll address use cases w.r.t. the Process API, shouldn't it? Kind regards, Anthony On Sunday, March 14, 2021 13:01 CET, Alan Bateman wrote: > On 14/03/2021 11:00, Marco wrote: > > : > > > > IMO Charset should provide standardized getters for the OS charset and the > > console charset. The latter being different has been a long standing issue > > on > > Windows where the codepage differs between its CLI and regular environments. > > OpenJDK has the necessary data already available in its custom system > > properties. > > > > The console charset is currently hidden behind PrintStream not exposing the > > underlying OSWriter and not offering getEncoding() itself. The OS charset > > would be hidden in the future by Charset.getDefaultCharset()'s specification > > change in JEP 400. > The intention that there will be little or no impact to the console > streams. This means that java.io.Console reader/writer methods should > continue to return a Reader/PrintWriter that uses the platform encoding > (or code page is on Windows). Same thing for the System.out/System.err > print streams. We need to make this clearer in the JEP. > > There has been discussion on this mailing list about adding a > Console::charset method but it didn't come to a consensus. Naoto Sato > and I have been chatting about it again recently as there may be a need > to add an API in advance of proposing to target the JEP. > > One case that we are still mulling over is code that creates an > InputStreamReader on System.in without specifying the charset. This > might be older code that pre-dates java.io.Console or maybe code that > wasn't tested on a wide range or platforms. Options range from a spec > change to doing nothing (the latter meaning running with "COMPACT" or > migrating the code to use the 2-arg constructor as the default charset > is not the right choice). > > -Alan > > >
Insufficiencies in JEP: 400: UTF-8 by Default
Hi all, the JEP generally paints the picture that using the OS charset would be incorrect or useless, it is however the right and perfectly valid choice for communicating with other local programs where no other charset was specified. It is the same as UTF-8 most of the time, but not always and especially not on Windows, using UTF-8 every time would be strictly less correct. Per [1] LC_CTYPE defines the charset to use for transforming between binary data and text. Given that the file.encoding system property doesn't exist within Java SE, LC_CTYPE combined with the current specification of Charset.defaultCharset() is the only compliant way to change the default charset in Java SE outside some custom application specific handling. Ignoring LC_CTYPE obviously leaves no standard approach. From the program's POV the same applies in reverse, currently one could only use Charset.defaultCharset() to determine the OS charset or let the java.io methods infer it through the charset-less constructors, then potentially read it back through e.g. InputStreamReader.getEncoding(). The OS charset is still relevant for text interaction on System.in/out/err, sub-process stdin/stdout/stderr and files with unknown encoding. Programs like grep assume the files are encoded according to LC_CTYPE, much like a similarly designed Java program that uses the OS charset on purpose. Constructing a Reader for stdin properly requires some way to determine the relevant OS encoding. I'm perfectly happy with changing the charset-less methods to use UTF-8 since it's the best choice outside the above scenarios, despite the compatibility impact. Dropping standardized support for the OS charset however not only breaks the above interactions, but also leaves no nice migration path. The - Dfile.encoding=COMPAT workaround is explicitly not standardized and isn't available to the Java application itself, only to whoever starts the JVM to presumably work around outdated code. IMO Charset should provide standardized getters for the OS charset and the console charset. The latter being different has been a long standing issue on Windows where the codepage differs between its CLI and regular environments. OpenJDK has the necessary data already available in its custom system properties. The console charset is currently hidden behind PrintStream not exposing the underlying OSWriter and not offering getEncoding() itself. The OS charset would be hidden in the future by Charset.getDefaultCharset()'s specification change in JEP 400. Please consider the above minor additions to fix those issues for good. Best regards, Marco [1] https://pubs.opengroup.org/onlinepubs/7908799/xbd/envvar.html
Re: Insufficiencies in JEP: 400: UTF-8 by Default
On 14/03/2021 11:00, Marco wrote: : IMO Charset should provide standardized getters for the OS charset and the console charset. The latter being different has been a long standing issue on Windows where the codepage differs between its CLI and regular environments. OpenJDK has the necessary data already available in its custom system properties. The console charset is currently hidden behind PrintStream not exposing the underlying OSWriter and not offering getEncoding() itself. The OS charset would be hidden in the future by Charset.getDefaultCharset()'s specification change in JEP 400. The intention that there will be little or no impact to the console streams. This means that java.io.Console reader/writer methods should continue to return a Reader/PrintWriter that uses the platform encoding (or code page is on Windows). Same thing for the System.out/System.err print streams. We need to make this clearer in the JEP. There has been discussion on this mailing list about adding a Console::charset method but it didn't come to a consensus. Naoto Sato and I have been chatting about it again recently as there may be a need to add an API in advance of proposing to target the JEP. One case that we are still mulling over is code that creates an InputStreamReader on System.in without specifying the charset. This might be older code that pre-dates java.io.Console or maybe code that wasn't tested on a wide range or platforms. Options range from a spec change to doing nothing (the latter meaning running with "COMPACT" or migrating the code to use the 2-arg constructor as the default charset is not the right choice). -Alan