bdbaddog / dummy-for-testing-migration Goto Github PK

This issue was originally created at: 2001-09-12 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-12 22:00:00

We probably shouldn't tackle this until after all of the other build issues are coded.

issues@scons said at 2001-09-12 22:00:00

Converted from SourceForge task item 37982

stevenknight said at 2006-05-20 20:51:52

No white space in keyword.

This issue was originally created at: 2001-09-12 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-12 22:00:00

Read options from $SCONSFLAGS in the external environment, and then let the command-line flags override them as appropriate. This must be documented at the same time the code goes in.

issues@scons said at 2001-09-12 22:00:00

Converted from SourceForge task item 37995

stevenknight said at 2006-05-20 20:51:15

No white space in keyword.

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

Actually search for #include dependencies using CPPPATH.

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 38993

stevenknight said at 2006-05-20 20:52:31

No white space in keyword.

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

Add support for the Repository() method and the -Y command-line option.

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 39012

This issue was originally created at: 2001-07-30 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-07-30 22:00:00

We need a coherent error handling architecture that distinguishes between user errors and internal errors.

issues@scons said at 2001-07-30 22:00:00

Converted from SourceForge task item 35383

stevenknight said at 2006-05-20 20:50:08

No white space in keyword.

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

This "prints internal environments/objects" a la the Gnu Make dump of its internal database. This should actually incorporate whatever's defined for local defaults, too, so the user can see what the system administrator or buildmeister has provided for them. Maybe it includes commentary, too, to explain what's what.

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 39024

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

Build targets/dependencies in random order.

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 39027

This issue was originally created at: 2001-07-19 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-07-19 22:00:00

Use the new Job class to support a simple end-to-end -j test.

issues@scons said at 2001-07-25 22:00:00

Converted from SourceForge task item 34796

stevenknight said at 2006-05-20 20:48:13

No white space in keyword.

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

Traverse up the directory tree looking for the SConstruct file. Rewrite targets as appropriate to build only in (and under) the local directory.

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 39028

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

Don't rebuild a file even if it needs it.

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 39023

This issue was originally created at: 2005-07-15 04:46:12.
This issue was reported by: gregnoel.
gregnoel said at 2005-07-15 04:46:12

This patch adds a FLAGS keyword to override environments (only). A complete
patch would also add a FLAGS keyword to standard environments, but that turned
out to be much more complex and I don't have time to do it before I leave on
holiday. I'm submitting this patch as a placeholder and a promise: if I make it
back (travel is getting more dangerous these days) I'll update this patch to
include standard environments. Worst case, someone else can do it. (Or someone
can work on it while I'm gone; fixing standard environments looks very messy, so
any help would be appreciated.)

What this patch allows is something like this:
env.Program('foo', 'bar.cpp', FLAGS = '-O2 -Dfoo -lfoobar')
and the flags will be automatically merged into the appropriate environment
variables. It invokes MergeFlags under the covers, so this will also work:
dict = env.ParseFlags('-O2 -Dfoo -lfoobar')
dict['SWIGFLAGS'] = '-c++ -python'
env.Object('bar.i', FLAGS = dict)
The flags can also be a list, with the expected semantics:
env.Program('foo', 'bar.cpp', FLAGS = ['-O2', '!pkg-config ...'])
Unfortunately, mixing dicts and strings isn't supported (maybe that will be a
future enhancement).

As a workaround for standard environments, this works:
env = Environment().MergeFlags(['-O2', '!pkg-config ...'])

stevenknight said at 2006-01-05 18:16:06

Changing version to "unspecified" to conform to new scheme.

kmaples said at 2006-05-20 17:03:29

making the default milestone consistent across the project

kmaples said at 2006-05-20 17:14:13

making the version consistent across the project

stevenknight said at 2006-05-26 03:46:34

Created an attachment (id=5)
diff for Environment.py

stevenknight said at 2006-05-26 03:47:47

Added file attachment as part of final severing of SourceForge ties.

gregnoel said at 2007-05-17 17:53:54

This should be assigned to me. I've already got some additional work beyond the
current patch and it's something I'd like to do.

gregnoel said at 2007-05-17 17:55:53

Humpf. I thought accepting an issue automatically assigned it to me.

gregnoel said at 2007-05-17 17:57:35

Okay, already; I guess I have to assign it to myself, then accept it.

gregnoel said at 2008-03-16 00:40:52

Created an attachment (id=331)
Final patch; applied 14-Mar-2008

gregnoel said at 2008-03-16 00:43:08

Patch applied; includes code, tests, docs, release description

More information about this issue is at http://simtech.sf.net/flags.txt.

gregnoel attached flags-diff at 2008-03-16 00:40:51.

Final patch; applied 14-Mar-2008

stevenknight attached flags.txt at 2008-03-25 11:08:15.

diff for Environment.py

This issue was originally created at: 2001-07-18 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-07-18 22:00:00

Get user documentation to the point where someone will be able to use the initial release to compile simple C programs.

issues@scons said at 2001-07-18 22:00:00

Converted from SourceForge task item 34779

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

We may not need this, since we don't chdir() around the tree the way recursive Make does. If we do need this, then we also need the --no-print-directory option.

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 39029

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

Support the ability to cut down the targets being built. The Cons method is a little quirky, so we should revisit whether or not it's the best way.

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 39014

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

This is a Cons classic option that we might not need if we have other, more GNU-like ways to do the same thing.

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 39030

This issue was originally created at: 2001-09-12 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-12 22:00:00

Get builds working reliably first, then we'll figure out how to ignore the errors.

issues@scons said at 2001-09-12 22:00:00

Converted from SourceForge task item 37987

stevenknight said at 2006-05-20 20:51:28

No white space in keyword.

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

Add something that lets the user print out signatures, with varying degrees of verboseness. We need this for development as much as users need this to know why SCons does what it does.

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 39013

This issue was originally created at: 2001-07-18 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-07-18 22:00:00

Requirements sent via email.

issues@scons said at 2001-07-18 22:00:00

Converted from SourceForge task item 34780

stevenknight said at 2006-05-20 20:47:53

No white space in keyword.

This issue was originally created at: 2001-09-20 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-20 22:00:00

Start collecting ideas on how we'll actually release the packages.

issues@scons said at 2001-09-20 22:00:00

Converted from SourceForge task item 38817

stevenknight said at 2006-05-20 20:40:17

Removing obsolete keyword.

This issue was originally created at: 2001-07-30 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-07-30 22:00:00

Be able to link in specified libraries using an LDFLAGS construction variable.

issues@scons said at 2001-07-30 22:00:00

Converted from SourceForge task item 35381

stevenknight said at 2006-05-20 20:50:37

No white space in keyword.

This issue was originally created at: 2001-07-19 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-07-19 22:00:00

Step-by-step for getting started.

issues@scons said at 2001-07-19 22:00:00

Converted from SourceForge task item 34801

stevenknight said at 2006-05-20 20:39:47

Removing obsolete keyword.

stevenknight said at 2007-02-07 21:54:16

We're moving towards basing everything off Subversion.

This issue was originally created at: 2005-06-16 13:49:58.
This issue was reported by: gregnoel.
gregnoel said at 2005-06-16 13:49:58

Expand framework support to compile and link steps (currently, frameworks are
just supported for bundles). Splits FRAMEWORKSFLAGS into two separate pieces:
FRAMEWORKPATH and FRAMEWORKS; both are a list of names. Created a construction
variable _CCCOMCOM to hold the C Compiler COMmon COMmand flags; _FRAMEWORKPATH
is appended to this for OS X. Psuedo-variables _FRAMEWORKPATH and _FRAMEWORKS
created to place the correct expansion of frameworks in the right places. Some
logic cleaned up: redundant code removed from gnulink.py and a typo fixed (the
rpath option has only one hyphen, not two).

I seem to have little success when trying to attach files, so the patch is in
the URL.

gregnoel said at 2005-06-16 13:52:21

Oops, sorry, this was supposed to be a patch, not a defect.

stevenknight said at 2006-01-05 18:15:24

Changing version to "unspecified" to conform to new scheme.

stevenknight said at 2006-05-03 18:45:42

Patch was integrated for release 0.96.91. Thank you!

kmaples said at 2006-05-20 17:03:29

making the default milestone consistent across the project

kmaples said at 2006-05-20 17:14:12

making the version consistent across the project

More information about this issue is at http://simtech.sf.net/diffs.txt.

This issue was originally created at: 2001-07-19 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-07-19 22:00:00

Tracks dependencies on in-core Python values. Dictionaries should be equivalent to directories, strings should be "files." This will allow us to write build-engine dependency tests more efficiently, without having to set up real files.

issues@scons said at 2001-07-19 22:00:00

Converted from SourceForge task item 34793

This issue was originally created at: 2001-07-19 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-07-19 22:00:00

Take the nodes returned from one or more Object() calls and feed them to Program() to create the final executable.

issues@scons said at 2001-07-19 22:00:00

Converted from SourceForge task item 34795

stevenknight said at 2006-05-20 20:49:26

No white space in keyword.

This issue was originally created at: 2001-09-12 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-12 22:00:00

Get builds working reliably, then we'll figure out how to keep going.

issues@scons said at 2001-09-12 22:00:00

Converted from SourceForge task item 37988

stevenknight said at 2006-05-20 20:52:04

No white space in keyword.

This issue was originally created at: 2001-07-30 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-07-30 22:00:00

Be able to install and use SCons with minimal hassle, whether from .rpm, .deb, or straight from .tar.gz.

issues@scons said at 2001-07-30 22:00:00

Converted from SourceForge task item 35382

stevenknight said at 2006-05-20 20:50:23

No white space in keyword.

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

The extremely cool GNU what-if option, showing what would be built if a file was changed. We should be able to do this easily by just adding a random (or hard-coded) value to the signature.

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 39031

This issue was originally created at: 2001-07-19 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-07-19 22:00:00

Be able to generate .o files explicitly.

issues@scons said at 2001-07-19 22:00:00

Converted from SourceForge task item 34794

stevenknight said at 2006-05-20 20:49:39

No white space in keyword.

This issue was originally created at: 2001-07-19 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-07-19 22:00:00

Tracking explicit dependencies will be easiest first.

issues@scons said at 2001-07-19 22:00:00

Converted from SourceForge task item 34792

stevenknight said at 2006-05-20 20:49:15

No white space in keyword.

This issue was originally created at: 2001-07-19 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-07-19 22:00:00

Step-by-step, covering how we're integrating this with Aegis under the hood.

issues@scons said at 2001-07-19 22:00:00

Converted from SourceForge task item 34802

stevenknight said at 2006-05-20 20:39:59

Removing obsolete keyword.

This issue was originally created at: 2001-07-19 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-07-19 22:00:00

This is where we add storing signature values from invocation to invocation in a .consign or .sconsign file.

issues@scons said at 2001-07-19 22:00:00

Converted from SourceForge task item 34799

stevenknight said at 2006-05-20 20:49:02

No white space in keyword.

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

A test prototype already exists in test/Library.py.

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 38980

stevenknight said at 2006-05-20 20:53:26

No white space in keyword.

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

From email from Knight, 24 Sept 2001:

Vague outline: A directory's entries should probably be entries in the sources[] attribute. Or better yet, turn that attribute into a _sources[] attribute, and use an accessor sources() method to get at it. The the Dir class can override sources() to just return its entries[] (minus '.' and '..').

A directory should always be "out of date" so that we always descend the tree and try to build the entries therein. I think this could be done by never giving it a real value for its signature, in which case the collected signatures of its entries would always be non-null and indicate the directory would need to be "rebuilt," but not before we've built its "source" entires.

The actual directory builder, then, would be just an internal do-nothing builder that updates status to indicate the directory has now been "built" (that is, all of its entries have been built).

A directory should never have a signature written to a .sconsign file.

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 38970

stevenknight said at 2006-05-20 20:52:55

No white space in keyword.

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

Support searching for libraries with which to link in the variables in LIBPATH.

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 38994

stevenknight said at 2006-05-20 20:52:18

No white space in keyword.

This issue was originally created at: 2001-07-19 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-07-19 22:00:00

The Builder class has explicit checks for the type of object used to build the target. It would be better OO design to do these as subclasses.

issues@scons said at 2001-07-19 22:00:00

Converted from SourceForge task item 34797

stevenknight said at 2006-05-20 20:48:48

No white space in keyword.

This issue was originally created at: 2006-01-24 01:33:19.
This issue was reported by: shlomif.
shlomif said at 2006-01-24 01:33:19

The search mailing list archive:

http://xrl.us/jper

Does not work regardless of what query one inputs there.

Please fix it.

Regards,

  Shlomi Fish

kmaples said at 2006-05-20 17:03:30

making the default milestone consistent across the project

kmaples said at 2006-05-20 17:14:13

making the version consistent across the project

stevenknight said at 2006-12-08 15:55:12

Not sure what mailing list archive that is, but it doesn't look like ours.

More information about this issue is at http://scons.tigris.org/servlets/SearchList?list=users&searchText=find&defaultField=body&Search=Search.

This issue was originally created at: 2005-08-11 14:34:17.
This issue was reported by: gregnoel.
gregnoel said at 2005-08-11 14:34:17

If a GCC compile command line has one or more '-include foo' options, no
dependency is created for the file 'foo'. Although the use of this option is
rare and tends to be to unchanging files, it's disconcerting to change such a
base file and not have the dependent files recompiled.

stevenknight said at 2006-01-05 18:16:19

Changing version to "unspecified" to conform to new scheme.

kmaples said at 2006-05-20 17:03:31

making the default milestone consistent across the project

kmaples said at 2006-05-20 17:14:14

making the version consistent across the project

bdbaddog said at 2008-03-15 22:43:18

Is scons supposed to be scanning the command line, or is it just supposed to use
CPPPATH to determine include paths?

gregnoel said at 2008-03-15 23:03:14

SCons frequently scans command lines to make decisions about what to do. For
example, the SWIG builder will generate different output depending upon whether
the -python or -c++ flag is present. This is no different.

Since it can lead to incorrect builds if the dependency isn't created, which
SCons will go to great lengths to avoid.

stevenknight said at 2008-10-02 10:42:18

See issue #1107, which we could possibly dup this to.

I'd prefer to see this configured by an explicit variable like $CPPFORCEINCLUDES
that then gets added to the command line with the (configurable) "-include"
option. This is more like the way we handle many other things, and allows the
user to be explicit about what .h file is used for the dependency.

Greg pointed out in IRC that doing so would have impacts on MergeFlags() and the
like, which makes sense.

I'll update #1107 with a comment. Greg, your call on whether we can go ahead
and dup this.

gregnoel said at 2008-10-02 16:34:32

Issue 1107 is the better approach, and it preceeds this one by five months, so
it gets the nod.

I don't like the name CPPFORCEINCLUDES, but I'll make that comment over there.

*** This issue has been marked as a duplicate of 1107 ***

gregnoel said at 2009-02-06 06:25:34

Bug party triage. Reopen this issue to add a CPPINCLUDE variable that contains
the list of headers to be used with -include on the command line; also add
CPPINCLUDE{PRE,SUF}FIX variables to expand to the correct wrapping. Names (and
Files) in this list should be dependencies of the compile step. See issue 1107
for a patch that does most of this.

stevenknight said at 2009-02-22 07:41:44

Assigning to SK and target milestone per bug party triage.

stevenknight said at 2009-11-10 18:00:20

stevenknight => issues@scons

gregnoel said at 2009-12-09 12:06:40

*** Issue 2475 has been marked as a duplicate of this issue. ***

Votes for this issue: 10.

gregnoel said this issue is duplicated by #2475 at 2009-12-09 12:06:41.

gregnoel said this issue is duplicated by #2475 at 2009-12-09 12:06:41.

This issue was originally created at: 2001-07-19 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-07-19 22:00:00

The purely Aegis "t0001.{py,t}" test names aren't very inviting. Give them real names (we're not going to care about order). Break up t0001.t while we're at it.

issues@scons said at 2001-07-19 22:00:00

Converted from SourceForge task item 34798

stevenknight said at 2006-05-20 20:48:30

No white space in keyword.

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

Allow compilation with different values of CCFLAGS. A prototype test already exists in test/CCFLAGS.py.

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 38992

stevenknight said at 2006-05-20 20:53:12

No white space in keyword.

This issue was originally created at: 2001-08-28 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-08-28 22:00:00

Through distutils, of course, although it will require some tweaking to start creating our own world, not just put things under the default C:\Python20 directory.

issues@scons said at 2001-08-28 22:00:00

Converted from SourceForge task item 37154

stevenknight said at 2006-05-20 20:50:49

No white space in keyword.

This issue was originally created at: 2005-06-05 08:02:45.
This issue was reported by: gregnoel.
gregnoel said at 2005-06-05 08:02:45

On May 10, 2005, at 13:01 US/Pacific, Steven Knight wrote, "... the current
behavior of AppendUnique() is sub-optimal and should be changed." In another
message, Gary Oberbrunner concured. The behavior in question is that
AppendUnique does not remove multiple occurences of the same value in the list
that is being appended. That is, if the new list is
['foo', 'bar', 'foobar', 'foo', 'bar']
all five elements are added, even though 'foo' and 'bar' are duplicated.

stevenknight said at 2006-01-05 18:15:11

Changing version to "unspecified" to conform to new scheme.

kmaples said at 2006-05-20 17:03:31

making the default milestone consistent across the project

kmaples said at 2006-05-20 17:14:14

making the version consistent across the project

bdbaddog said at 2008-03-15 22:40:58

Does the following, demonstrate the bug?

env=Environment(tool=[])
env['TEST3']=['a']
env.AppendUnique(TEST3=['a','a','b','b','a','c','c'])
print 'TEST3 is now:%s'%env['TEST3']

Which outputs (in latest svn):
TEST3 is now:['a', 'b', 'b', 'c', 'c']

gregnoel said at 2008-11-30 12:24:54

Bug party triage. May already be fixed. Gary to research.

garyo said at 2008-12-04 17:40:24

Still broken in current trunk. Will fix.

garyo said at 2008-12-04 19:23:07

Fixed in r3804.

evgen said at 2016-08-30 17:50:48

Not resolved, according to this discussion:
http://four.pairlist.net/pipermail/scons-users/2013-September/001736.html

• at least, when nodes are involved

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

Clear out (don't use) default environment/objects. These meant different things to Gnu Make, but they're probably the same for us. Or should we separate them somehow?

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 39026

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

Support specification of libraries with which we'll link Program() files in a LIBS construction variable.

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 38995

stevenknight said at 2006-05-20 20:52:43

No white space in keyword.

This issue was originally created at: 2001-09-12 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-12 22:00:00

After we have dependencies coded.

issues@scons said at 2001-09-12 22:00:00

Converted from SourceForge task item 37985

This issue was originally created at: 2001-08-28 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-08-28 22:00:00

Make sure we can create .deb files as well. Ideally, this would be done through distutils, but that didn't support .deb natively the last time I looked.

issues@scons said at 2001-08-28 22:00:00

Converted from SourceForge task item 37153

stevenknight said at 2006-05-20 20:51:02

No white space in keyword.

This issue was originally created at: 2001-09-12 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-12 22:00:00

Change to '$' as the interpolation character, and modification flags that look like Python methods, per the mailing list discussion.

issues@scons said at 2001-09-12 22:00:00

Converted from SourceForge task item 37993

stevenknight said at 2006-05-20 20:51:40

No white space in keyword.

This issue was originally created at: 2001-07-30 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-07-30 22:00:00

Create a Scanner object that looks for #include lines in C/C++ files.

issues@scons said at 2001-07-30 22:00:00

Converted from SourceForge task item 35419

stevenknight said at 2006-05-20 20:49:51

No white space in keyword.

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

Pick up all the grungy registry logic that we'll need to make this transparent from distutils.

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 39033

This issue was originally created at: 2001-09-23 22:00:00.
This issue was reported by: stevenknight.
stevenknight said at 2001-09-23 22:00:00

A GNU Make option. Python will throw exceptions for us if an undefined Python variable is used in the script portion itself, but our construction variable expansion should support a way to do this as well.

issues@scons said at 2001-09-23 22:00:00

Converted from SourceForge task item 39032

This issue was originally created at: 2005-07-15 03:59:28.
This issue was reported by: gregnoel.
gregnoel said at 2005-07-15 03:59:28

If a source file (such as a header) is modified in mid-build, the signature of
the modified file is retained in the dependencies for targets already built
using the unmodified file. A subsequent run of SCons should rebuild those
components that used the unmodified file, but does not.

This was acknowledged as a bug by Steven Knight when it was reported on the
mailing list and is recorded here so that the issue does not get lost.

stevenknight said at 2006-01-05 18:15:54

Changing version to "unspecified" to conform to new scheme.

kmaples said at 2006-05-20 17:03:28

making the default milestone consistent across the project

kmaples said at 2006-05-20 17:14:12

making the version consistent across the project

stevenknight said at 2007-05-22 08:39:58

This needs a specific test case (or more than one) to isolate the specific
problem(s) being described.

If the header file is a generated file, then anything that uses that header file
better have a dependency on it, or else the DAG is incomplete. If the DAG is
complete, then the header file would be updated before anything that uses it.

If the problem here is that something unknown to SCons can modify a header file
mid-build and mess up the signatures, the eventual solution there is some sort
of "validation" mode that would check every time a file is referenced to make
sure (e.g.) its timestamp is still the same as the original reference.

Please update (and/or open new issues) to clarify.

issues@scons said at 2007-05-22 14:25:38

Created an attachment (id=190)
Zip file to demonstrate bug

gregnoel said at 2007-05-22 14:47:14

The bug that I was originally reporting seems to be fixed (the saved signature
reflects the contents of the file when it is first encountered instead of when
it is last encountered), but it does leave the strangeness identified in the
zipfile immediately prior to this note. You have to decide if it's a feature or
a bug.

Unzip the file into a scratch directory. You should get a SConstruct, a shell
script, a header file, and two C files.

Run 'scons' the first time. All three build steps should be executed. What has
happened is that the middle build step has changed the header file, so the first
build is now out-of-date. (The third is OK since the header was changed before
it ran.)

Run 'scons' again. The middle build step is up-to-date, so it is not run, but
both of the other two steps are run. This is correct for one.c, since hdr.h has
changed since it was last run, but redundant for two.c, since hdr.h has not
changed since it was last run.

(In previous releases, none of the build steps would be run, since the sig for
hdr.h was saved for the run of two.c and used for one.c. Now it seems to save
the sig at the first reference and use it for both.)

Run 'scons' a third time. This time everything is up-to-date and it correctly
does nothing.

Failing to rebuild one.c is clearly a bug and seems to have gone away. An extra
rebuild of two.c is a waste of cycles, but otherwise harmless. It's your choice
as to whether this situation is still a bug. If you think it's OK, go ahead and
close this issue.

gregnoel said at 2007-05-27 11:12:57

*** Issue 6 has been confirmed by votes. ***

stevenknight said at 2007-08-29 17:48:45

I'm not going to fix this--at least now--but I've captured the test case in a
script so that it can be modified if we ever change this behavior--or used as
the basis for a companion script if we add some configurable mode or the like:

http://scons.tigris.org/servlets/ReadMsg?list=dev&msgNo=4210
http://scons.tigris.org/servlets/ReadMsg?list=commits&msgNo=965

If someone else wants to take a crack at it, feel free to open this back up. It
will involve examining every file for every built node to make sure it hasn't
changed (an N^2 issue when lots of .h files are involved), and copying the
NodeInfo objects that store each file's metadata (content, signature,
timestamp), instead of just holding a list of pointers to each Node's own
NodeInfo object.

An anonymous user attached bug6.zip at 2007-05-22 14:25:37.

Zip file to demonstrate bug