JIT works only partially with meson build?

Started by Yugo Nagata8 months ago6 messages

nagata@sraoss.co.jp

8 months ago

Hi,

While building PostgreSQL 17 on Windows, I noticed bitcode files (.bc) are not
generated with meson. Does this means that "inlining" of JIT doesn't
work when PostgreSQL is build with meson?

If so, is it better to describe that in the meson build and JIT sections in the
documentation so that users and packagers would except JIT doesn't work in the same
way when built with meson and autoconf/make?

Regards,
Yugo Nagata

--
Yugo Nagata <nagata@sraoss.co.jp>

dgrowleyml@gmail.com

8 months ago

In reply to: Yugo Nagata (#1)

Re: JIT works only partially with meson build?

On Mon, 26 May 2025 at 18:08, Yugo Nagata <nagata@sraoss.co.jp> wrote:

While building PostgreSQL 17 on Windows, I noticed bitcode files (.bc) are not
generated with meson. Does this means that "inlining" of JIT doesn't
work when PostgreSQL is build with meson?

Yes. I believe [1]https://commitfest.postgresql.org/patch/5671/ aims to fix this.

David

[1]: https://commitfest.postgresql.org/patch/5671/

nagata@sraoss.co.jp

7 months ago

In reply to: David Rowley (#2)

1 attachment(s)

Re: JIT works only partially with meson build?

On Mon, 26 May 2025 19:17:53 +1200
David Rowley <dgrowleyml@gmail.com> wrote:

On Mon, 26 May 2025 at 18:08, Yugo Nagata <nagata@sraoss.co.jp> wrote:

While building PostgreSQL 17 on Windows, I noticed bitcode files (.bc) are not
generated with meson. Does this means that "inlining" of JIT doesn't
work when PostgreSQL is build with meson?

Yes. I believe [1] aims to fix this.

David

[1] https://commitfest.postgresql.org/patch/5671/

Thank you for the clarification.

I've attached a patch to update the documentation.
I believe it's worth stating this explicitly, since the fix will not be
applied until at the earliest PostgreSQL 19.

Regards,
Yugo Nagata

--
Yugo Nagata <nagata@sraoss.co.jp>

Attachments:

fix_doc_jit_meson.patchtext/x-diff; name=fix_doc_jit_meson.patchDownload

diff --git a/doc/src/sgml/jit.sgml b/doc/src/sgml/jit.sgml
index 44e18bf1a6f..00092675288 100644
--- a/doc/src/sgml/jit.sgml
+++ b/doc/src/sgml/jit.sgml
@@ -36,7 +36,8 @@
    <acronym>JIT</acronym> compilation using <ulink
    url="https://llvm.org/"><productname>LLVM</productname></ulink> when
    <productname>PostgreSQL</productname> is built with
-   <link linkend="configure-with-llvm"><literal>--with-llvm</literal></link>.
+   <link linkend="configure-with-llvm"><literal>--with-llvm</literal></link>
+   or <link linkend="configure-with-llvm-meson"><option>-Dllvm</option></link>.
   </para>
 
   <para>
@@ -76,6 +77,15 @@
     bodies of small functions into the expressions using them. That allows a
     significant percentage of the overhead to be optimized away.
    </para>
+   <note>
+    <para>
+    Inlining does not work if <productname>PostgreSQL</productname> is built using
+    <application>Meson</application> with
+    <link linkend="configure-with-llvm-meson"><option>-Dllvm</option></link>,
+    because the <productname>LLVM</productname> bitcode is not built.
+    See <xref linkend="jit-extensibility-bitcode"/> for more details about bitcode.
+    </para>
+   </note>
   </sect2>
 
   <sect2 id="jit-optimization">

laurenz.albe@cybertec.at

3 months ago

In reply to: Yugo Nagata (#3)

Re: JIT works only partially with meson build?

On Fri, 2025-06-27 at 16:19 +0900, Yugo Nagata wrote:

On Mon, 26 May 2025 19:17:53 +1200
David Rowley <dgrowleyml@gmail.com> wrote:

On Mon, 26 May 2025 at 18:08, Yugo Nagata <nagata@sraoss.co.jp> wrote:

While building PostgreSQL 17 on Windows, I noticed bitcode files (.bc) are not
generated with meson. Does this means that "inlining" of JIT doesn't
work when PostgreSQL is build with meson?

Yes. I believe [1] aims to fix this.

[1] https://commitfest.postgresql.org/patch/5671/

Thank you for the clarification.

I've attached a patch to update the documentation.
I believe it's worth stating this explicitly, since the fix will not be
applied until at the earliest PostgreSQL 19.

I think that it is a good idea to document this lack of functionality.

One thing stood out to me:

--- a/doc/src/sgml/jit.sgml
+++ b/doc/src/sgml/jit.sgml
@@ -36,7 +36,8 @@
<acronym>JIT</acronym> compilation using <ulink
url="https://llvm.org/"><productname>LLVM</productname></ulink> when
<productname>PostgreSQL</productname> is built with
-   <link linkend="configure-with-llvm"><literal>--with-llvm</literal></link>.
+   <link linkend="configure-with-llvm"><literal>--with-llvm</literal></link>
+   or <link linkend="configure-with-llvm-meson"><option>-Dllvm</option></link>.

You used the <option> tag, which I believe is correct, but the surrounding
code uses <literal>. I don't think it renders differently, but I think it
would be good to use the same tag. I suggest that you change the
surrounding code to use <option> as well.

v18 is already shipped, and we don't know if the patch to produce bitcode
with Meson will make v19 or not. Consequently, I don't know if we should
apply the patch to git HEAD yet. But I believe that it should be
backpatched to all older releases, since this is a shortcoming in the
documentation.

Yours,
Laurenz Albe

nagata@sraoss.co.jp

3 months ago

In reply to: Laurenz Albe (#4)

1 attachment(s)

Re: JIT works only partially with meson build?

On Tue, 30 Sep 2025 14:04:13 +0200
Laurenz Albe <laurenz.albe@cybertec.at> wrote:

On Fri, 2025-06-27 at 16:19 +0900, Yugo Nagata wrote:

On Mon, 26 May 2025 19:17:53 +1200
David Rowley <dgrowleyml@gmail.com> wrote:

On Mon, 26 May 2025 at 18:08, Yugo Nagata <nagata@sraoss.co.jp> wrote:

While building PostgreSQL 17 on Windows, I noticed bitcode files (.bc) are not
generated with meson. Does this means that "inlining" of JIT doesn't
work when PostgreSQL is build with meson?

Yes. I believe [1] aims to fix this.

[1] https://commitfest.postgresql.org/patch/5671/

Thank you for the clarification.

I've attached a patch to update the documentation.
I believe it's worth stating this explicitly, since the fix will not be
applied until at the earliest PostgreSQL 19.

I think that it is a good idea to document this lack of functionality.

Thank you for taking a look at it.

One thing stood out to me:
--- a/doc/src/sgml/jit.sgml
+++ b/doc/src/sgml/jit.sgml
@@ -36,7 +36,8 @@
<acronym>JIT</acronym> compilation using <ulink
url="https://llvm.org/"><productname>LLVM</productname></ulink> when
<productname>PostgreSQL</productname> is built with
-   <link linkend="configure-with-llvm"><literal>--with-llvm</literal></link>.
+   <link linkend="configure-with-llvm"><literal>--with-llvm</literal></link>
+   or <link linkend="configure-with-llvm-meson"><option>-Dllvm</option></link>.
You used the <option> tag, which I believe is correct, but the surrounding
code uses <literal>. I don't think it renders differently, but I think it
would be good to use the same tag. I suggest that you change the
surrounding code to use <option> as well.

Agreed. I fixed it to use <option>, as well as in other place where I think
<option> is appropriate. Please find the attached patch.

v18 is already shipped, and we don't know if the patch to produce bitcode
with Meson will make v19 or not. Consequently, I don't know if we should
apply the patch to git HEAD yet. But I believe that it should be
backpatched to all older releases, since this is a shortcoming in the
documentation.

+1.

Regards,
Yugo Nagata

--
Yugo Nagata <nagata@sraoss.co.jp>

Attachments:

v2_fix_doc_jit_meson.patchtext/x-diff; name=v2_fix_doc_jit_meson.patchDownload

diff --git a/doc/src/sgml/jit.sgml b/doc/src/sgml/jit.sgml
index 44e18bf1a6f..d5373061e3a 100644
--- a/doc/src/sgml/jit.sgml
+++ b/doc/src/sgml/jit.sgml
@@ -36,7 +36,8 @@
    <acronym>JIT</acronym> compilation using <ulink
    url="https://llvm.org/"><productname>LLVM</productname></ulink> when
    <productname>PostgreSQL</productname> is built with
-   <link linkend="configure-with-llvm"><literal>--with-llvm</literal></link>.
+   <link linkend="configure-with-llvm"><option>--with-llvm</option></link>
+   or <link linkend="configure-with-llvm-meson"><option>-Dllvm</option></link>.
   </para>
 
   <para>
@@ -76,6 +77,15 @@
     bodies of small functions into the expressions using them. That allows a
     significant percentage of the overhead to be optimized away.
    </para>
+   <note>
+    <para>
+    Inlining does not work if <productname>PostgreSQL</productname> is built using
+    <application>Meson</application> with
+    <link linkend="configure-with-llvm-meson"><option>-Dllvm</option></link>,
+    because the <productname>LLVM</productname> bitcode is not built.
+    See <xref linkend="jit-extensibility-bitcode"/> for more details about bitcode.
+    </para>
+   </note>
   </sect2>
 
   <sect2 id="jit-optimization">
@@ -133,7 +143,7 @@
    <para>
     If <xref linkend="guc-jit"/> is set to <literal>off</literal>, or if no
     <acronym>JIT</acronym> implementation is available (for example because
-    the server was compiled without <literal>--with-llvm</literal>),
+    the server was compiled without <option>--with-llvm</option>),
     <acronym>JIT</acronym> will not be performed, even if it would be
     beneficial based on the above criteria.  Setting <xref linkend="guc-jit"/>
     to <literal>off</literal> has effects at both plan and execution time.

laurenz.albe@cybertec.at

3 months ago

In reply to: Yugo Nagata (#5)

Re: JIT works only partially with meson build?

On Tue, 2025-09-30 at 22:31 +0900, Yugo Nagata wrote:

You used the <option> tag, which I believe is correct, but the surrounding
code uses <literal>. I don't think it renders differently, but I think it
would be good to use the same tag. I suggest that you change the
surrounding code to use <option> as well.

Agreed. I fixed it to use <option>, as well as in other place where I think
<option> is appropriate. Please find the attached patch.

I'll mark it "eeady for committer". A committer can decide whether to apply
this to HEAD or not. I probably would, and revert it if PostgreSQL + Meson +
Windows learns to produce bitcode.

Yours,
Laurenz Albe