[PATCH] Carry the gnulib getcwd backport as a patch

[PATCH] Carry the gnulib getcwd backport as a patch

[Simon Marchi]

Commit

  e2fc52e7457 ("Fix PR gdb/23558: Use system's 'getcwd' when cross-compiling GDB")

backported some changes from a future gnulib version to our import.
However, this means that every time someone wants to change our gnulib
import (e.g. add a module), they must make sure not to include that
backported change.  It also means that someone running the
update-gnulib.sh script without changes will get some diffs and wonder
why.

Instead, I suggest we carry that backport as a patch applied by the
update-gnulib.sh script after running the import tool.  It will make it
clear what backport or local modification we have and should make
running update-gnulib.sh give a reproducible result.

There is a hunk in the configure file in this patch, this is because the
commit that backported the getcwd bits didn't include the re-generated
configure.

Note: you'll need this patch as well to get deterministic results:

  Generate aclocal-m4-deps.mk more deterministically and portably.
  https://sourceware.org/ml/gdb-patches/2018-09/msg00643.html

gdb/ChangeLog:

	* patches/0001-Fix-PR-gdb-23558-Use-system-s-getcwd-when-cross-comp.patch:
	New file.
	* update-gnulib.sh: Apply patch.
	* configure: Re-generate.
---
 gdb/gnulib/configure                          | 18 ++++-
 ...-Use-system-s-getcwd-when-cross-comp.patch | 69 +++++++++++++++++++
 gdb/gnulib/update-gnulib.sh                   | 12 ++++
 3 files changed, 96 insertions(+), 3 deletions(-)
 create mode 100644 gdb/gnulib/patches/0001-Fix-PR-gdb-23558-Use-system-s-getcwd-when-cross-comp.patch

diff --git a/gdb/gnulib/configure b/gdb/gnulib/configure
index 8c3027635f5..092db16ba28 100644
--- a/gdb/gnulib/configure
+++ b/gdb/gnulib/configure
@@ -16921,10 +16921,22 @@ else
   # Arrange for deletion of the temporary directory this test creates.
      ac_clean_files="$ac_clean_files confdir3"
           if test "$cross_compiling" = yes; then :
-  case "$host_os" in
-       aix*) gl_cv_func_getcwd_path_max='no, it has the AIX bug';;
-       *) gl_cv_func_getcwd_path_max=no;;
+  # Cross-compilation guesses:
+     case "$host_os" in
+       aix*) # On AIX, it has the AIX bug.
+         gl_cv_func_getcwd_path_max='no, it has the AIX bug' ;;
+       gnu*) # On Hurd, it is 'yes'.
+         gl_cv_func_getcwd_path_max=yes ;;
+       linux* | kfreebsd*)
+         # On older Linux+glibc it's 'no, but it is partly working',
+         # on newer Linux+glibc it's 'yes'.
+         # On Linux+musl libc, it's 'no, but it is partly working'.
+         # On kFreeBSD+glibc, it's 'no, but it is partly working'.
+         gl_cv_func_getcwd_path_max='no, but it is partly working' ;;
+       *) # If we don't know, assume the worst.
+         gl_cv_func_getcwd_path_max=no ;;
      esac
+
 else
   cat confdefs.h - <<_ACEOF >conftest.$ac_ext
 /* end confdefs.h.  */
diff --git a/gdb/gnulib/patches/0001-Fix-PR-gdb-23558-Use-system-s-getcwd-when-cross-comp.patch b/gdb/gnulib/patches/0001-Fix-PR-gdb-23558-Use-system-s-getcwd-when-cross-comp.patch
new file mode 100644
index 00000000000..1d8cdf6b901
--- /dev/null
+++ b/gdb/gnulib/patches/0001-Fix-PR-gdb-23558-Use-system-s-getcwd-when-cross-comp.patch
@@ -0,0 +1,69 @@
+From a7447215cdc492a077732420a3f0db6d48b34d51 Mon Sep 17 00:00:00 2001
+From: Sergio Durigan Junior <sergiodj@redhat.com>
+Date: Mon, 10 Sep 2018 12:52:04 -0400
+Subject: [PATCH] Fix PR gdb/23558: Use system's 'getcwd' when cross-compiling
+ GDB
+
+This is a backport of a gnulib fix for the following bug:
+
+  https://sourceware.org/bugzilla/show_bug.cgi?id=23558
+
+The problem reported there is about the replacement of 'getcwd' when
+cross-compiling GDB.  With our current gnulib copy, the mechanism for
+deciding whether to use the system's 'getcwd' or gnulib's version is
+too simplistic and pessimistic, so when cross-compiling we always end
+up using gnulib's version, which has a limitation: it cannot handle
+the situation when the parent directory doesn't have read permissions.
+
+The solution is to backport the following gnulib commit:
+
+  commit a96d2e67052c879b1bcc5bc461722beac75fc372
+  Author: Bruno Haible <bruno@clisp.org>
+  Date:   Thu Aug 23 21:13:19 2018 +0200
+
+      getcwd: Add cross-compilation guesses.
+
+gdb/ChangeLog:
+2018-09-10  Sergio Durigan Junior  <sergiodj@redhat.com>
+
+	PR gdb/23555
+	PR gdb/23558
+	* gnulib/import/m4/getcwd-path-max.m4: Add cross-compilation
+	guesses.
+---
+ gdb/gnulib/import/m4/getcwd-path-max.m4 | 20 ++++++++++++++++----
+ 1 file changed, 16 insertions(+), 4 deletions(-)
+
+diff --git a/gdb/gnulib/import/m4/getcwd-path-max.m4 b/gdb/gnulib/import/m4/getcwd-path-max.m4
+index 2531ccff65c..90bbc77dea1 100644
+--- a/gdb/gnulib/import/m4/getcwd-path-max.m4
++++ b/gdb/gnulib/import/m4/getcwd-path-max.m4
+@@ -209,9 +209,21 @@ main ()
+      32) gl_cv_func_getcwd_path_max='yes, but with shorter paths';;
+      *) gl_cv_func_getcwd_path_max=no;;
+      esac],
+-    [case "$host_os" in
+-       aix*) gl_cv_func_getcwd_path_max='no, it has the AIX bug';;
+-       *) gl_cv_func_getcwd_path_max=no;;
+-     esac])
++    [# Cross-compilation guesses:
++     case "$host_os" in
++       aix*) # On AIX, it has the AIX bug.
++         gl_cv_func_getcwd_path_max='no, it has the AIX bug' ;;
++       gnu*) # On Hurd, it is 'yes'.
++         gl_cv_func_getcwd_path_max=yes ;;
++       linux* | kfreebsd*)
++         # On older Linux+glibc it's 'no, but it is partly working',
++         # on newer Linux+glibc it's 'yes'.
++         # On Linux+musl libc, it's 'no, but it is partly working'.
++         # On kFreeBSD+glibc, it's 'no, but it is partly working'.
++         gl_cv_func_getcwd_path_max='no, but it is partly working' ;;
++       *) # If we don't know, assume the worst.
++         gl_cv_func_getcwd_path_max=no ;;
++     esac
++    ])
+   ])
+ ])
+-- 
+2.19.0
+
diff --git a/gdb/gnulib/update-gnulib.sh b/gdb/gnulib/update-gnulib.sh
index 3077424622b..dff1254830b 100755
--- a/gdb/gnulib/update-gnulib.sh
+++ b/gdb/gnulib/update-gnulib.sh
@@ -157,6 +157,18 @@ if [ $? -ne 0 ]; then
    exit 1
 fi
 
+# Apply our local patches.
+apply_patches ()
+{
+    patch -p3 -f -i "$1"
+    if [ $? -ne 0 ]; then
+        echo "Failed to apply some patches.  Aborting."
+        exit 1
+    fi
+}
+
+apply_patches "patches/0001-Fix-PR-gdb-23558-Use-system-s-getcwd-when-cross-comp.patch"
+
 # Regenerate all necessary files...
 aclocal -Iimport/m4 &&
 autoconf &&
-- 
2.19.0

Re: [PATCH] Carry the gnulib getcwd backport as a patch

[John Baldwin]

On 9/18/18 2:21 PM, Simon Marchi wrote:
> Commit
> 
>   e2fc52e7457 ("Fix PR gdb/23558: Use system's 'getcwd' when cross-compiling GDB")
> 
> backported some changes from a future gnulib version to our import.
> However, this means that every time someone wants to change our gnulib
> import (e.g. add a module), they must make sure not to include that
> backported change.  It also means that someone running the
> update-gnulib.sh script without changes will get some diffs and wonder
> why.
> 
> Instead, I suggest we carry that backport as a patch applied by the
> update-gnulib.sh script after running the import tool.  It will make it
> clear what backport or local modification we have and should make
> running update-gnulib.sh give a reproducible result.
> 
> There is a hunk in the configure file in this patch, this is because the
> commit that backported the getcwd bits didn't include the re-generated
> configure.

This looks good to me.  I've verified it generates ok for me on a FreeBSD host
as well.

-- 
John Baldwin

                                                                            

Re: [PATCH] Carry the gnulib getcwd backport as a patch

[Tom Tromey]

>>>>> "Simon" == Simon Marchi <simon.marchi@ericsson.com> writes:

Simon> Instead, I suggest we carry that backport as a patch applied by the
Simon> update-gnulib.sh script after running the import tool.  It will make it
Simon> clear what backport or local modification we have and should make
Simon> running update-gnulib.sh give a reproducible result.

This seems like a good idea to me.

Simon> +# Apply our local patches.
Simon> +apply_patches ()

This should probably be "apply_patch" since it just applies a single patch.

Also there is a comment near the top of update-gnulib.sh:

    # This script will also make a number of other verifications:
    #   . The gnulib version (it should match $GNULIB_COMMIT_SHA1).
    #   . The correct versions of the auto-tools that are used to
    #     regenerate the various scripts and Makefiles are on the PATH.

I think an addition here about updating the patches would be good.

Tom

Re: [PATCH] Carry the gnulib getcwd backport as a patch

[Simon Marchi]

On 2018-09-18 05:38 PM, Tom Tromey wrote:
> Simon> +# Apply our local patches.
> Simon> +apply_patches ()
> 
> This should probably be "apply_patch" since it just applies a single patch.

Oops yes.

> Also there is a comment near the top of update-gnulib.sh:
> 
>     # This script will also make a number of other verifications:
>     #   . The gnulib version (it should match $GNULIB_COMMIT_SHA1).
>     #   . The correct versions of the auto-tools that are used to
>     #     regenerate the various scripts and Makefiles are on the PATH.
> 
> I think an addition here about updating the patches would be good.

I don't think it belongs in that particular list, because it's not a verification,
but I've added a note in that header.

Does this look good?


From f0584664629ee49216e5b56810c9eec83f9f3a8f Mon Sep 17 00:00:00 2001
From: Simon Marchi <simon.marchi@ericsson.com>
Date: Tue, 18 Sep 2018 16:42:53 -0400
Subject: [PATCH] Carry the gnulib getcwd backport as a patch

Commit

  e2fc52e7457 ("Fix PR gdb/23558: Use system's 'getcwd' when cross-compiling GDB")

backported some changes from a future gnulib version to our import.
However, this means that every time someone wants to change our gnulib
import (e.g. add a module), they must make sure not to include that
backported change.  It also means that someone running the
update-gnulib.sh script without changes will get some diffs and wonder
why.

Instead, I suggest we carry that backport as a patch applied by the
update-gnulib.sh script after running the import tool.  It will make it
clear what backport or local modification we have and should make
running update-gnulib.sh give a reproducible result.

There is a hunk in the configure file in this patch, this is because the
commit that backported the getcwd bits didn't include the re-generated
configure.

gdb/ChangeLog:

	* patches/0001-Fix-PR-gdb-23558-Use-system-s-getcwd-when-cross-comp.patch:
	New file.
	* update-gnulib.sh: Apply patch.
	* configure: Re-generate.
---
 gdb/gnulib/configure                          | 18 ++++-
 ...-Use-system-s-getcwd-when-cross-comp.patch | 69 +++++++++++++++++++
 gdb/gnulib/update-gnulib.sh                   | 14 ++++
 3 files changed, 98 insertions(+), 3 deletions(-)
 create mode 100644 gdb/gnulib/patches/0001-Fix-PR-gdb-23558-Use-system-s-getcwd-when-cross-comp.patch

diff --git a/gdb/gnulib/configure b/gdb/gnulib/configure
index 5fd52d45ac9..dc6c5b66573 100644
--- a/gdb/gnulib/configure
+++ b/gdb/gnulib/configure
@@ -17696,10 +17696,22 @@ else
   # Arrange for deletion of the temporary directory this test creates.
      ac_clean_files="$ac_clean_files confdir3"
           if test "$cross_compiling" = yes; then :
-  case "$host_os" in
-       aix*) gl_cv_func_getcwd_path_max='no, it has the AIX bug';;
-       *) gl_cv_func_getcwd_path_max=no;;
+  # Cross-compilation guesses:
+     case "$host_os" in
+       aix*) # On AIX, it has the AIX bug.
+         gl_cv_func_getcwd_path_max='no, it has the AIX bug' ;;
+       gnu*) # On Hurd, it is 'yes'.
+         gl_cv_func_getcwd_path_max=yes ;;
+       linux* | kfreebsd*)
+         # On older Linux+glibc it's 'no, but it is partly working',
+         # on newer Linux+glibc it's 'yes'.
+         # On Linux+musl libc, it's 'no, but it is partly working'.
+         # On kFreeBSD+glibc, it's 'no, but it is partly working'.
+         gl_cv_func_getcwd_path_max='no, but it is partly working' ;;
+       *) # If we don't know, assume the worst.
+         gl_cv_func_getcwd_path_max=no ;;
      esac
+
 else
   cat confdefs.h - <<_ACEOF >conftest.$ac_ext
 /* end confdefs.h.  */
diff --git a/gdb/gnulib/patches/0001-Fix-PR-gdb-23558-Use-system-s-getcwd-when-cross-comp.patch b/gdb/gnulib/patches/0001-Fix-PR-gdb-23558-Use-system-s-getcwd-when-cross-comp.patch
new file mode 100644
index 00000000000..1d8cdf6b901
--- /dev/null
+++ b/gdb/gnulib/patches/0001-Fix-PR-gdb-23558-Use-system-s-getcwd-when-cross-comp.patch
@@ -0,0 +1,69 @@
+From a7447215cdc492a077732420a3f0db6d48b34d51 Mon Sep 17 00:00:00 2001
+From: Sergio Durigan Junior <sergiodj@redhat.com>
+Date: Mon, 10 Sep 2018 12:52:04 -0400
+Subject: [PATCH] Fix PR gdb/23558: Use system's 'getcwd' when cross-compiling
+ GDB
+
+This is a backport of a gnulib fix for the following bug:
+
+  https://sourceware.org/bugzilla/show_bug.cgi?id=23558
+
+The problem reported there is about the replacement of 'getcwd' when
+cross-compiling GDB.  With our current gnulib copy, the mechanism for
+deciding whether to use the system's 'getcwd' or gnulib's version is
+too simplistic and pessimistic, so when cross-compiling we always end
+up using gnulib's version, which has a limitation: it cannot handle
+the situation when the parent directory doesn't have read permissions.
+
+The solution is to backport the following gnulib commit:
+
+  commit a96d2e67052c879b1bcc5bc461722beac75fc372
+  Author: Bruno Haible <bruno@clisp.org>
+  Date:   Thu Aug 23 21:13:19 2018 +0200
+
+      getcwd: Add cross-compilation guesses.
+
+gdb/ChangeLog:
+2018-09-10  Sergio Durigan Junior  <sergiodj@redhat.com>
+
+	PR gdb/23555
+	PR gdb/23558
+	* gnulib/import/m4/getcwd-path-max.m4: Add cross-compilation
+	guesses.
+---
+ gdb/gnulib/import/m4/getcwd-path-max.m4 | 20 ++++++++++++++++----
+ 1 file changed, 16 insertions(+), 4 deletions(-)
+
+diff --git a/gdb/gnulib/import/m4/getcwd-path-max.m4 b/gdb/gnulib/import/m4/getcwd-path-max.m4
+index 2531ccff65c..90bbc77dea1 100644
+--- a/gdb/gnulib/import/m4/getcwd-path-max.m4
++++ b/gdb/gnulib/import/m4/getcwd-path-max.m4
+@@ -209,9 +209,21 @@ main ()
+      32) gl_cv_func_getcwd_path_max='yes, but with shorter paths';;
+      *) gl_cv_func_getcwd_path_max=no;;
+      esac],
+-    [case "$host_os" in
+-       aix*) gl_cv_func_getcwd_path_max='no, it has the AIX bug';;
+-       *) gl_cv_func_getcwd_path_max=no;;
+-     esac])
++    [# Cross-compilation guesses:
++     case "$host_os" in
++       aix*) # On AIX, it has the AIX bug.
++         gl_cv_func_getcwd_path_max='no, it has the AIX bug' ;;
++       gnu*) # On Hurd, it is 'yes'.
++         gl_cv_func_getcwd_path_max=yes ;;
++       linux* | kfreebsd*)
++         # On older Linux+glibc it's 'no, but it is partly working',
++         # on newer Linux+glibc it's 'yes'.
++         # On Linux+musl libc, it's 'no, but it is partly working'.
++         # On kFreeBSD+glibc, it's 'no, but it is partly working'.
++         gl_cv_func_getcwd_path_max='no, but it is partly working' ;;
++       *) # If we don't know, assume the worst.
++         gl_cv_func_getcwd_path_max=no ;;
++     esac
++    ])
+   ])
+ ])
+--
+2.19.0
+
diff --git a/gdb/gnulib/update-gnulib.sh b/gdb/gnulib/update-gnulib.sh
index 327b37f3a3c..90b103e50a2 100755
--- a/gdb/gnulib/update-gnulib.sh
+++ b/gdb/gnulib/update-gnulib.sh
@@ -27,6 +27,8 @@
 #   . The gnulib version (it should match $GNULIB_COMMIT_SHA1).
 #   . The correct versions of the auto-tools that are used to
 #     regenerate the various scripts and Makefiles are on the PATH.
+#
+# It will also apply any local patches to gnulib we carry at the time.

 # The list of gnulib modules we are importing in GDB.
 IMPORTED_GNULIB_MODULES="\
@@ -158,6 +160,18 @@ if [ $? -ne 0 ]; then
    exit 1
 fi

+# Apply our local patches.
+apply_patch()
+{
+    patch -p3 -f -i "$1"
+    if [ $? -ne 0 ]; then
+        echo "Failed to apply some patches.  Aborting."
+        exit 1
+    fi
+}
+
+apply_patch "patches/0001-Fix-PR-gdb-23558-Use-system-s-getcwd-when-cross-comp.patch"
+
 # Regenerate all necessary files...
 aclocal -Iimport/m4 &&
 autoconf &&
-- 
2.19.0

Re: [PATCH] Carry the gnulib getcwd backport as a patch

[Simon Marchi]

On 2018-09-18 17:55, Tom Tromey wrote:
> Simon> Does this look good?
> 
> Yes, thanks.
> 
> Tom

Thanks, I'll wait to see if Sergio has any comment about this before 
pushing.

Simon

Re: [PATCH] Carry the gnulib getcwd backport as a patch

[Sergio Durigan Junior]

On Tuesday, September 18 2018, Simon Marchi wrote:

> On 2018-09-18 17:55, Tom Tromey wrote:
>> Simon> Does this look good?
>>
>> Yes, thanks.
>>
>> Tom
>
> Thanks, I'll wait to see if Sergio has any comment about this before
> pushing.

Thanks for taking care of this.  I agree with the approach taken here
(as discussed on IRC), and am OK with the patch.

-- 
Sergio
GPG key ID: 237A 54B1 0287 28BF 00EF  31F4 D0EB 7628 65FC 5E36
Please send encrypted e-mail if possible
http://sergiodj.net/

Re: [PATCH] Carry the gnulib getcwd backport as a patch

[Simon Marchi]

On 2018-09-18 22:42, Sergio Durigan Junior wrote:
> On Tuesday, September 18 2018, Simon Marchi wrote:
> 
>> On 2018-09-18 17:55, Tom Tromey wrote:
>>> Simon> Does this look good?
>>> 
>>> Yes, thanks.
>>> 
>>> Tom
>> 
>> Thanks, I'll wait to see if Sergio has any comment about this before
>> pushing.
> 
> Thanks for taking care of this.  I agree with the approach taken here
> (as discussed on IRC), and am OK with the patch.

Thanks, I pushed it.

Simon