Hi, all. Consider the following command, as run on a Mac with an up-to-date Xcode installed:
xcrun swiftc foo.swift
The question: should this build for the current running OS, or the oldest macOS Swift supports (10.9)? You can always specify the deployment target OS version explicitly with the -target option, but what should the default behavior be?
Some points to consider:
- The deployment OS affects availability checks, which means that the command might succeed on one host but fail on another.
- …but we already changed the default for the interpreter (`xcrun swift`) to be the current running OS in Swift 3.1 (Xcode 8.3, last spring).
- Clang defaults to the current running OS (as of a few Xcodes ago, IIRC).
Given these points, I'm inclined to change swiftc to default to building for the current running OS when no target is specified, but what do other people think?
Note that this doesn't apply to projects built with either Xcode or the Swift Package Manager, both of which always explicitly provide a deployment target. Invoking swiftc directly and not providing -target means (1) you are definitely compiling for Mac (when run on a Mac), and (2) there's a good chance you don't plan to distribute what you just built, because until Swift lives in the OS, it has dependencies on your installed Swift toolchain (currently messily resolved with absolute rpaths). If you avoid this with -static-stdlib, you're giving up the ability to have dynamic libraries, because we didn't implement that properly.
Thanks for your feedback!
Jordan
P.S. For Apple folks, this is rdar://problem/29948658.
It would be my assumption that it would build for the currently running OS. It would be very confusing to me to have availability checks failing for an OS version lower that what I'm using. I'm sure that I would be swearing for a few hours before I finally found the obscure documentation that said that it would compile for the oldest OS swift supports.
Thanks for asking us! :)
- Will
···
On Nov 27, 2017, at 4:44 PM, Jordan Rose via swift-dev <swift-dev@swift.org> wrote:
Hi, all. Consider the following command, as run on a Mac with an up-to-date Xcode installed:
xcrun swiftc foo.swift
The question: should this build for the current running OS, or the oldest macOS Swift supports (10.9)? You can always specify the deployment target OS version explicitly with the -target option, but what should the default behavior be?
Some points to consider:
- The deployment OS affects availability checks, which means that the command might succeed on one host but fail on another.
- …but we already changed the default for the interpreter (`xcrun swift`) to be the current running OS in Swift 3.1 (Xcode 8.3, last spring).
- Clang defaults to the current running OS (as of a few Xcodes ago, IIRC).
Given these points, I'm inclined to change swiftc to default to building for the current running OS when no target is specified, but what do other people think?
Note that this doesn't apply to projects built with either Xcode or the Swift Package Manager, both of which always explicitly provide a deployment target. Invoking swiftc directly and not providing -target means (1) you are definitely compiling for Mac (when run on a Mac), and (2) there's a good chance you don't plan to distribute what you just built, because until Swift lives in the OS, it has dependencies on your installed Swift toolchain (currently messily resolved with absolute rpaths). If you avoid this with -static-stdlib, you're giving up the ability to have dynamic libraries, because we didn't implement that properly.
Thanks for your feedback!
Jordan
P.S. For Apple folks, this is rdar://problem/29948658 <rdar://problem/29948658>.
On Nov 27, 2017, at 16:44, Jordan Rose via swift-dev <swift-dev@swift.org> wrote:
Hi, all. Consider the following command, as run on a Mac with an up-to-date Xcode installed:
xcrun swiftc foo.swift
The question: should this build for the current running OS, or the oldest macOS Swift supports (10.9)? You can always specify the deployment target OS version explicitly with the -target option, but what should the default behavior be?
Some points to consider:
- The deployment OS affects availability checks, which means that the command might succeed on one host but fail on another.
- …but we already changed the default for the interpreter (`xcrun swift`) to be the current running OS in Swift 3.1 (Xcode 8.3, last spring).
- Clang defaults to the current running OS (as of a few Xcodes ago, IIRC).
Given these points, I'm inclined to change swiftc to default to building for the current running OS when no target is specified, but what do other people think?
Note that this doesn't apply to projects built with either Xcode or the Swift Package Manager, both of which always explicitly provide a deployment target. Invoking swiftc directly and not providing -target means (1) you are definitely compiling for Mac (when run on a Mac), and (2) there's a good chance you don't plan to distribute what you just built, because until Swift lives in the OS, it has dependencies on your installed Swift toolchain (currently messily resolved with absolute rpaths). If you avoid this with -static-stdlib, you're giving up the ability to have dynamic libraries, because we didn't implement that properly.
Thanks for your feedback!
Jordan
P.S. For Apple folks, this is rdar://problem/29948658 <rdar://problem/29948658>.
I agree, the currently running OS seems like the right default here.
Progressive disclosure and ease of prototyping are good motivations here.
If I just want to quickly prototype something, I'm not going to be thinking
about choosing a minimum OS; I'm just going to write something using the
APIs that are available on my current one. If I decide to distribute that
later, *that's* when I'm going to start thinking about minimums (and adding
availability directives, if needed).
···
On Mon, Nov 27, 2017 at 4:53 PM William Dillon via swift-dev < swift-dev@swift.org> wrote:
It would be my assumption that it would build for the currently running
OS. It would be very confusing to me to have availability checks failing
for an OS version lower that what I'm using. I'm sure that I would be
swearing for a few hours before I finally found the obscure documentation
that said that it would compile for the oldest OS swift supports.
Thanks for asking us! :)
- Will
On Nov 27, 2017, at 4:44 PM, Jordan Rose via swift-dev < > swift-dev@swift.org> wrote:
Hi, all. Consider the following command, as run on a Mac with an
up-to-date Xcode installed:
xcrun swiftc foo.swift
The question: should this build for the *current running OS,* or *the
oldest macOS Swift supports* (10.9)? You can always specify the
deployment target OS version explicitly with the -target option, but what
should the default behavior be?
Some points to consider:
- The deployment OS affects availability checks, which means that the
command might succeed on one host but fail on another.
- …but we already changed the default for the interpreter (`xcrun swift`)
to be the current running OS in Swift 3.1 (Xcode 8.3, last spring).
- Clang defaults to the current running OS (as of a few Xcodes ago, IIRC).
Given these points, I'm inclined to change swiftc to default to building
for the current running OS when no target is specified, but what do other
people think?
Note that this doesn't apply to projects built with either Xcode or the
Swift Package Manager, both of which always explicitly provide a deployment
target. Invoking swiftc directly and not providing -target means (1) you
are definitely compiling for Mac (when run on a Mac), and (2) there's a
good chance you don't plan to distribute what you just built, because until
Swift lives in the OS, it has dependencies on your installed Swift
toolchain (currently messily resolved with absolute rpaths). If you avoid
this with -static-stdlib, you're giving up the ability to have dynamic
libraries, because we didn't implement that properly.
Thanks for your feedback!
Jordan
P.S. For Apple folks, this is rdar://problem/29948658.
I think you'll still have to write the availability checks, because Swift will make you. I think the fixit also tells you what you need for the API you're trying to use. The confusing thing will be, though, that it'll use the fall-back path unless you explicitly set the target version.
···
On Nov 28, 2017, at 10:36 AM, Tony Allevato via swift-dev <swift-dev@swift.org> wrote:
I agree, the currently running OS seems like the right default here. Progressive disclosure and ease of prototyping are good motivations here. If I just want to quickly prototype something, I'm not going to be thinking about choosing a minimum OS; I'm just going to write something using the APIs that are available on my current one. If I decide to distribute that later, *that's* when I'm going to start thinking about minimums (and adding availability directives, if needed).
On Mon, Nov 27, 2017 at 4:53 PM William Dillon via swift-dev <swift-dev@swift.org <mailto:swift-dev@swift.org>> wrote:
It would be my assumption that it would build for the currently running OS. It would be very confusing to me to have availability checks failing for an OS version lower that what I'm using. I'm sure that I would be swearing for a few hours before I finally found the obscure documentation that said that it would compile for the oldest OS swift supports.
Thanks for asking us! :)
- Will
On Nov 27, 2017, at 4:44 PM, Jordan Rose via swift-dev <swift-dev@swift.org <mailto:swift-dev@swift.org>> wrote:
Hi, all. Consider the following command, as run on a Mac with an up-to-date Xcode installed:
xcrun swiftc foo.swift
The question: should this build for the current running OS, or the oldest macOS Swift supports (10.9)? You can always specify the deployment target OS version explicitly with the -target option, but what should the default behavior be?
Some points to consider:
- The deployment OS affects availability checks, which means that the command might succeed on one host but fail on another.
- …but we already changed the default for the interpreter (`xcrun swift`) to be the current running OS in Swift 3.1 (Xcode 8.3, last spring).
- Clang defaults to the current running OS (as of a few Xcodes ago, IIRC).
Given these points, I'm inclined to change swiftc to default to building for the current running OS when no target is specified, but what do other people think?
Note that this doesn't apply to projects built with either Xcode or the Swift Package Manager, both of which always explicitly provide a deployment target. Invoking swiftc directly and not providing -target means (1) you are definitely compiling for Mac (when run on a Mac), and (2) there's a good chance you don't plan to distribute what you just built, because until Swift lives in the OS, it has dependencies on your installed Swift toolchain (currently messily resolved with absolute rpaths). If you avoid this with -static-stdlib, you're giving up the ability to have dynamic libraries, because we didn't implement that properly.
Thanks for your feedback!
Jordan
P.S. For Apple folks, this is rdar://problem/29948658 <>.
There should not be any availability checks required in the common fast-prototyping case, where your SDK version and your running OS version are the same.
Availability checks are required when some API has changed since the OS version that is your deployment target. In the fast-prototyping scenario the SDK describes no OS versions newer than the default deployment target. Therefore no such API changes are possible.
As Tony said, if you subsequently set the deployment target to be older or build with the same deployment target using a newer SDK then you may need to write availability checks.
···
On Nov 28, 2017, at 11:15 AM, William Dillon via swift-dev <swift-dev@swift.org> wrote:
I think you'll still have to write the availability checks, because Swift will make you. I think the fixit also tells you what you need for the API you're trying to use. The confusing thing will be, though, that it'll use the fall-back path unless you explicitly set the target version.
On Nov 28, 2017, at 10:36 AM, Tony Allevato via swift-dev <swift-dev@swift.org <mailto:swift-dev@swift.org>> wrote:
I agree, the currently running OS seems like the right default here. Progressive disclosure and ease of prototyping are good motivations here. If I just want to quickly prototype something, I'm not going to be thinking about choosing a minimum OS; I'm just going to write something using the APIs that are available on my current one. If I decide to distribute that later, *that's* when I'm going to start thinking about minimums (and adding availability directives, if needed).
On Mon, Nov 27, 2017 at 4:53 PM William Dillon via swift-dev <swift-dev@swift.org <mailto:swift-dev@swift.org>> wrote:
It would be my assumption that it would build for the currently running OS. It would be very confusing to me to have availability checks failing for an OS version lower that what I'm using. I'm sure that I would be swearing for a few hours before I finally found the obscure documentation that said that it would compile for the oldest OS swift supports.
Thanks for asking us! :)
- Will
On Nov 27, 2017, at 4:44 PM, Jordan Rose via swift-dev <swift-dev@swift.org <mailto:swift-dev@swift.org>> wrote:
Hi, all. Consider the following command, as run on a Mac with an up-to-date Xcode installed:
xcrun swiftc foo.swift
The question: should this build for the current running OS, or the oldest macOS Swift supports (10.9)? You can always specify the deployment target OS version explicitly with the -target option, but what should the default behavior be?
Some points to consider:
- The deployment OS affects availability checks, which means that the command might succeed on one host but fail on another.
- …but we already changed the default for the interpreter (`xcrun swift`) to be the current running OS in Swift 3.1 (Xcode 8.3, last spring).
- Clang defaults to the current running OS (as of a few Xcodes ago, IIRC).
Given these points, I'm inclined to change swiftc to default to building for the current running OS when no target is specified, but what do other people think?
Note that this doesn't apply to projects built with either Xcode or the Swift Package Manager, both of which always explicitly provide a deployment target. Invoking swiftc directly and not providing -target means (1) you are definitely compiling for Mac (when run on a Mac), and (2) there's a good chance you don't plan to distribute what you just built, because until Swift lives in the OS, it has dependencies on your installed Swift toolchain (currently messily resolved with absolute rpaths). If you avoid this with -static-stdlib, you're giving up the ability to have dynamic libraries, because we didn't implement that properly.
Thanks for your feedback!
Jordan
P.S. For Apple folks, this is rdar://problem/29948658 <>.
On Nov 28, 2017, at 10:36 AM, Tony Allevato via swift-dev <swift-dev@swift.org> wrote:
I agree, the currently running OS seems like the right default here. Progressive disclosure and ease of prototyping are good motivations here. If I just want to quickly prototype something, I'm not going to be thinking about choosing a minimum OS; I'm just going to write something using the APIs that are available on my current one. If I decide to distribute that later, *that's* when I'm going to start thinking about minimums (and adding availability directives, if needed).
On Mon, Nov 27, 2017 at 4:53 PM William Dillon via swift-dev <swift-dev@swift.org <mailto:swift-dev@swift.org>> wrote:
It would be my assumption that it would build for the currently running OS. It would be very confusing to me to have availability checks failing for an OS version lower that what I'm using. I'm sure that I would be swearing for a few hours before I finally found the obscure documentation that said that it would compile for the oldest OS swift supports.
Thanks for asking us! :)
- Will
On Nov 27, 2017, at 4:44 PM, Jordan Rose via swift-dev <swift-dev@swift.org <mailto:swift-dev@swift.org>> wrote:
Hi, all. Consider the following command, as run on a Mac with an up-to-date Xcode installed:
xcrun swiftc foo.swift
The question: should this build for the current running OS, or the oldest macOS Swift supports (10.9)? You can always specify the deployment target OS version explicitly with the -target option, but what should the default behavior be?
Some points to consider:
- The deployment OS affects availability checks, which means that the command might succeed on one host but fail on another.
- …but we already changed the default for the interpreter (`xcrun swift`) to be the current running OS in Swift 3.1 (Xcode 8.3, last spring).
- Clang defaults to the current running OS (as of a few Xcodes ago, IIRC).
Given these points, I'm inclined to change swiftc to default to building for the current running OS when no target is specified, but what do other people think?
Note that this doesn't apply to projects built with either Xcode or the Swift Package Manager, both of which always explicitly provide a deployment target. Invoking swiftc directly and not providing -target means (1) you are definitely compiling for Mac (when run on a Mac), and (2) there's a good chance you don't plan to distribute what you just built, because until Swift lives in the OS, it has dependencies on your installed Swift toolchain (currently messily resolved with absolute rpaths). If you avoid this with -static-stdlib, you're giving up the ability to have dynamic libraries, because we didn't implement that properly.
Thanks for your feedback!
Jordan
P.S. For Apple folks, this is rdar://problem/29948658 <>.